[go: up one dir, main page]

US20080050357A1 - Systems and Methods for Antibody Engineering - Google Patents

Systems and Methods for Antibody Engineering Download PDF

Info

Publication number
US20080050357A1
US20080050357A1 US10/566,954 US56695404A US2008050357A1 US 20080050357 A1 US20080050357 A1 US 20080050357A1 US 56695404 A US56695404 A US 56695404A US 2008050357 A1 US2008050357 A1 US 2008050357A1
Authority
US
United States
Prior art keywords
virus
antibody
variant
substitutions
sequence
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Abandoned
Application number
US10/566,954
Other languages
English (en)
Inventor
Claes Gustafsson
Sridhar Govindarajan
Jeremy Stephen Minshull
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
DNA Twopointo Inc
Original Assignee
DNA Twopointo Inc
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by DNA Twopointo Inc filed Critical DNA Twopointo Inc
Priority to US10/566,954 priority Critical patent/US20080050357A1/en
Assigned to DNA TWOPOINTO, INC. reassignment DNA TWOPOINTO, INC. ASSIGNMENT OF ASSIGNORS INTEREST (SEE DOCUMENT FOR DETAILS). Assignors: GUSTAFSSON, CLAES, GOVINDARAJAN, SRIDHAR, MINSHULL, JEREMY STEPHEN
Publication of US20080050357A1 publication Critical patent/US20080050357A1/en
Priority to US12/726,843 priority patent/US8412461B2/en
Priority to US13/854,002 priority patent/US20140032186A1/en
Abandoned legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • C—CHEMISTRY; METALLURGY
    • C07—ORGANIC CHEMISTRY
    • C07K—PEPTIDES
    • C07K16/00—Immunoglobulins [IGs], e.g. monoclonal or polyclonal antibodies
    • C07K16/08—Immunoglobulins [IGs], e.g. monoclonal or polyclonal antibodies against material from viruses
    • C07K16/10—Immunoglobulins [IGs], e.g. monoclonal or polyclonal antibodies against material from viruses from RNA viruses
    • C07K16/1027—Paramyxoviridae, e.g. respiratory syncytial virus
    • C—CHEMISTRY; METALLURGY
    • C07—ORGANIC CHEMISTRY
    • C07K—PEPTIDES
    • C07K16/00—Immunoglobulins [IGs], e.g. monoclonal or polyclonal antibodies
    • C07K16/46—Hybrid immunoglobulins
    • C07K16/461—Igs containing Ig-regions, -domains or -residues form different species
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B15/00—ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B15/00—ICT specially adapted for analysing two-dimensional or three-dimensional molecular structures, e.g. structural or functional relations or structure alignment
    • G16B15/20—Protein or domain folding
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/20—Allele or variant detection, e.g. single nucleotide polymorphism [SNP] detection
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B20/00—ICT specially adapted for functional genomics or proteomics, e.g. genotype-phenotype associations
    • G16B20/50—Mutagenesis
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
    • G16B40/30—Unsupervised data analysis
    • C—CHEMISTRY; METALLURGY
    • C07—ORGANIC CHEMISTRY
    • C07K—PEPTIDES
    • C07K2317/00—Immunoglobulins specific features
    • C07K2317/70—Immunoglobulins specific features characterized by effect upon binding to a cell or to an antigen
    • C07K2317/76—Antagonist effect on antigen, e.g. neutralization or inhibition of binding
    • G—PHYSICS
    • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
    • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
    • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding

Definitions

  • the field of this invention relates to computer systems and methods for designing sets of antibody variants and tools for relating the functional properties of such antibodies to their sequences. These relationships can then be used to determine the relationship between an antibody's sequence and commercially relevant properties of that antibody. Such sequence-function relationships may be used to design and synthesize commercially useful antibody compositions.
  • Empirical engineering of antibodies relies upon creating and testing sets of variants, then using this information to design and synthesize subsequent sets of variants that are enriched for components that contribute to the desired activity.
  • a key limitation for any empirical antibody engineering is in developing a good assay for antibody function.
  • the assay must measure antibody properties that are relevant to the final application, but must also be capable of testing a sufficient number of variants to identify what may be only a small fraction that are actually improved.
  • the difficulty of creating such an assay is particularly relevant when optimizing antibodies for complex functions that are difficult to measure in high throughput. Examples include reduction of viral titer or the killing of tumor cells.
  • Limitations in current methods for searching through antibody sequences for specific commercially relevant functionalities creates a need in the art for methods that can design and synthesize small numbers of variants for functional testing and that can use the resulting sequence and functional information to design and synthesize small numbers of variants improved for a desired commercially useful activity.
  • Limitations in current methods for choosing surrogate screens appropriate for empirical antibody engineering creates a need in the art for methods that can design and create small numbers of variants that can then be tested for specific commercially relevant functionalities.
  • each antibody is also designed to maximize the information that the set of antibodies contains regarding the contribution of substitutions to the desired antibody properties and to the contributions resulting from interactions between substitutions. This in essence is a map of the sequence space that can also be used to design perturbations to modify the functionality of the antibody as desired.
  • the information used to create the substitutions that define the sequence space can be derived from one or more of (i) multiple sequence alignments, (ii) phylogenetic reconstructions of ancestral sequences, (iii) analysis of families or superfamilies of antibodies related by sequence, structure, function or partial function, (iv) analysis of monomer substitution probabilities within classes of antibody, (v) three dimensional structures (e.g., molecular models, X-ray crystallographic structures, nuclear magnetic resonance models, molecular dynamic simulations), (vi) immunogenic constraints, (vii) prior knowledge about the structure and/or function of the sequences upon which design of the antibody set is to be based, or (viii) any similar information pertaining to a related or homologous antibody.
  • this process is automated by use of an expert system that acquires domain knowledge and captures it is a knowledge database.
  • This process can provide a score or rank order of substitutions to be incorporated, and a reasoning based on user specified constraints and domain specific data.
  • the first step in the design and manufacture of the statistically representative sequence sets of this invention is the definition of the initial sequence space to be searched. This involves defining one or more reference sequences, identifying positions that are likely to tolerate alteration, and identifying substitutions at these positions that are likely to be acceptable or to produce desired changes in the properties of the antibody. All possible combinatorial strings of polymeric biological molecules define the total defined sequence space to be searched. Each substitution at each position is typically enumerated in silico and the acceptability defined computationally. Desirability or acceptability of each possible substitution is calculated according to one or more criteria. Such calculations can be performed by a computational system using the knowledge database, user specified constraints, and/or domain and antibody specific data.
  • the present invention also provides a more formal systematic method for selecting substitution positions.
  • the use of a formal system involves quantitative scores and/or filters for assessing the favorability of substitution positions and the substitutions possible at those positions.
  • Formalizing the system for substitution selection allows for the development of an automated system for antibody optimization or humanization.
  • the parameters, filters and scores can be adjusted based on data from the scientific literature and data from experiments designed or interpreted by the automated system.
  • substitutions that are predicted to be favorable can be aligned with those found experimentally to be favorable.
  • Continuous refinement of these scores and filters based on experimental or computational data provides a way for the antibody optimization system to learn and improve. This formalization and learning capability are an aspect of the invention.
  • the second step in the design and manufacture of the statistically representative sequence sets of this invention is to define a subspace of the total sequence space to be searched in each iteration of the synthesis testing and correlating process.
  • the total allowed space matrix contains 10 5 -10 50 antibodies, many orders of magnitude larger than can be synthesized and measured under commercially relevant conditions. Such commercially relevant conditions are presently limited to numbers in the range of 10 1 -10 3 .
  • the number of antibody variants that can be synthesized and tested under appropriate conditions is defined by the availability of resources.
  • the number of variant positions and the number of substitutions that can be tested at each of those positions is then calculated, such that each substitution will be present in a statistically representative fraction of the set of antibodies to be synthesized.
  • search methods like Tabu, Ant optimization or similar techniques, the space can be searched on a sequence by sequence basis by using a memory of the space that has been visited previously and the properties encountered.
  • Typical experimental design methods can introduce more changes in an antibody than the antibody can tolerate to remain functional. Adaptations of these methods, for example by using covering algorithms to reduce the total number of substitutions in each antibody variant, while maximizing the number of different combinations of pairs of substitutions is another aspect of the invention.
  • the third step in the design and manufacture of the statistically representative sequence sets (or sequence sets relevant for specific optimization techniques) of this invention is to create a set of variant antibodies. This can be performed by synthesizing the antibody sequences defined and designed in the first two steps.
  • the systematic design of such variants is one aspect of the present invention.
  • the antibodies can be synthesized individually, or in a multiplexed set that is subsequently deconvoluted by sequencing or some other appropriate method. Alternatively, the antibodies can be created as a library of variants. Many methods have been described in the art for creating such libraries.
  • the designed set(s) of antibodies are characterized functionally to measure the properties of interest. This requires the development of an assay or surrogate assay faithful to the property or properties of ultimate interest and to test some members of the set of variants for more than one property, including the property of ultimate interest.
  • Data mining techniques are then employed to characterize the functions of the variants and to derive a relationship between antibody sequences and properties.
  • the characterization data can be used to provide information in a subsequent iteration of the method, aiding in the design of a subsequent set of statistically representative variants that can be synthesized and tested to obtain a molecule with even more desirable properties.
  • the data from additional iterations of this process can also be used to refine the data mining algorithms and models produced from the first set of data.
  • the knowledge created about the sequence space can in turn be incorporated into the knowledge database for evaluating the substitutions in the light of this data and recalculating the scores or rank order of the substitutions.
  • combinations of the methods described herein can be made with other techniques such as directed evolution, DNA shuffling, family shuffling and/or systematic scanning approaches. These can be performed in any order and for any number of iterations to produce the products described herein. All such combinations are within the scope of the invention.
  • FIG. 1 illustrates an overview of the architecture of an Expert System in accordance with an embodiment of the present invention.
  • FIG. 2 illustrates a flowchart for an antibody engineering method using integrated information sources to choose initial substitutions, and sequence-activity relationships to assess them in accordance with an embodiment of the present invention.
  • FIG. 3 is a schematic representation of a method for selecting amino acid substitutions for the optimization or humanization of antibodies in accordance with an embodiment of the present invention.
  • FIG. 4 illustrates a method for calculation of weights (e.g. contributions to activity) for each amino acid substitution in accordance with an embodiment of the present invention.
  • FIG. 5 illustrates a method for calculation of weights (e.g., contributions to activity) for each substitution in accordance with an embodiment of the present invention. This method provides information about the confidence of each weight by comparison with weights obtained from randomized data.
  • weights e.g., contributions to activity
  • FIG. 6 illustrates the amino acid sequence of wild type proteinase K, reported by Gunkel et al. (1989) Eur J Biochem 179: 185-194, modified by (i) replacement of the fungal leader peptide with an E. coli leader peptide, amino acids ⁇ 20 to ⁇ 1 (SEQ ID No. 1), and (ii) addition of a histidine tag to the C terminus (amino acids 372-377), together with a ValAsp preceding the tag (amino acids 370 and 371) to accommodate cloning sites in the nucleic acid sequence.
  • FIG. 7 illustrates the nucleotide sequence of proteinase K optimized for expression in E. coli .
  • the E. coli leader peptide (amino acids ⁇ 20 to ⁇ 1 in FIG. 6 ) are encoded by nucleotides ⁇ 60 to ⁇ 1 in FIG. 7 .
  • the proteinase K sequence beginning with Ala at amino acid 1 and ending with Ala at amino acid 369, is encoded by nucleotides 1-1107.
  • the histidine tag, the two additional amino acids described in FIG. 6 and the termination codon are encoded by nucleotides 1108-1133.
  • FIG. 8 shows the accession numbers of 49 proteinase K homologs obtained by BLAST searching of Genbank.
  • FIG. 9 illustrates a distribution of proteinase K homolog sequences (listed in FIG. 8 ) in the first two principal components of the sequence space. Sequences 46-49 are derived from thermostable organisms.
  • FIG. 10 illustrates a corresponding plot of all loads describing the influence of each variable on the sample distribution of FIG. 9
  • FIG. 11 provides magnified detail of the bottom left quadrant from FIG. 10 .
  • FIG. 12 provides principal component analysis-derived loads for individual amino acids responsible for clustering of thermostable proteinase K homologs.
  • FIG. 13 illustrates sample output from an Expert System defining the 24 most highly scoring substitutions to be incorporated into a set of variants for initial mapping of proteinase K sequence-function space in accordance with an embodiment of the present invention.
  • FIG. 14 illustrates a first designed set of 24 variants for proteinase K. Each variant contains six substitutions from the wild type sequence. The numbers refer to the substitutions identified in FIG. 13 .
  • FIG. 15 illustrates a second designed set of variants for proteinase K.
  • FIGS. 16A-16F illustrate amino acid changes in a set of synthesized proteinase K variants. Each column shows the changes from the wild type sequence present in one variant. A blank cell indicates the wild type sequence at that position. Amino acid numbering is shown in FIG. 6 .
  • FIGS. 17A and 17B provide activity measurements of proteinase K variants. Proteinase K variants were assessed for six different hydrolytic activities. All activities are normalized to the average performance of the wild type proteinase K.
  • y1 hydrolysis of a modified tetrapeptide, N-succinyl-Ala-Ala-Pro-Leu-p-nitroanilide (AAPL-p-NA) by purified proteinase K variants at pH 7.5
  • y2 thermostability ratio: activity after heat/activity without heat treatment, y6/y1
  • y4 hydrolysis of a modified tetrapeptide, N-succinyl-Ala-Ala-Pro-Leu-p-nitroanilide (AAPL-p-NA) by purified proteinase K variants at pH 4.5
  • y5 hydrolysis of a modified tetrapeptide, N-succinyl-Ala-Ala-Pro-
  • FIG. 18 illustrates a comparison between values predicted and values measured for a protein sequence-activity model derived from sequences shown in FIG. 16 and activity data (y6) shown in FIG. 17 .
  • Measured activities of proteinase K variant activities towards AAPL-p-NA following a five minute 65° C. heat treatment on the y-axis are compared with those predicted by the model on the x-axis. All activities were measured at 37° C. and pH 7.0 using purified protein.
  • FIG. 19 illustrates the identification of amino acids contributing to a specific function from a sequence-activity model. Regression coefficients (squares, left axis) of variant amino acids were derived from the sequence-activity model relating the sequences of proteinase K sequence variants (with numbers lower than 49) to activity y6. The number of occurrences of each amino acid substitution are also shown (diamonds, right axis). Changes from the wild type sequence are circled.
  • FIG. 20 illustrates the use of sequence-activity modeling to design a new variant with improved activity.
  • Four amino acid substitutions were found to have positive regression coefficients in their contribution to activity following heat-treatment (y6).
  • the variant test set contained one variant with one of these changes (#19) and one with three of these changes (#40).
  • a new variant (#56) was synthesized to contain all four changes.
  • the graph shows the activity of these variants towards AAPL-p-NA following five minute 65° C. heat treatment. Purified proteins were heated to 65° C. then incubated with AAPL-p-NA at pH 7.5. The reaction was followed by measuring the absorbance at 405 nm.
  • Alterations from the wild type sequence are: #19, K208H (filled triangles); #S40, V2671, G293A, K332R (open circles); #56, K208H, V2671, G293A, K332R (filled squares).
  • FIG. 21 illustrates how different amino acids are important for different functions in proteinase K. Beneficial amino acid substitutions were calculated by sequence-activity modeling for three different proteinase K properties. Changes from the wild type sequence are underlined.
  • FIG. 22 is a schematic representation of a method for selecting amino acid substitutions for the optimization of antiviral activity of an antibody in accordance with an embodiment of the present invention.
  • FIG. 23 is a schematic representation of a method for selecting amino acid substitutions for the humanization of an antibody in accordance with an embodiment of the present invention.
  • FIG. 24 is a list of germline sequence locus identification numbers obtained from VBase (http://www.mrc-cpe.cam.ac.uk)
  • FIG. 25 illustrates a distribution of RSV antibody and antibody sequences (listed in FIG. 24 ) in the first two principal components of the sequence space.
  • FIG. 26 illustrates a corresponding plot of all loads describing the influence of each variable on the sample distribution of FIG. 25
  • FIG. 27 provides magnified detail of the right center from FIG. 26
  • FIG. 28 provides principal component analysis-derived loads for individual amino acids responsible for clustering of sequences in group containing the sequence 4-28.
  • FIG. 29 is a list of germline sequence locus identification numbers obtained from VBase (http://www.mrc-cpe.cam.ac.uk)
  • FIG. 30 illustrates a distribution of AAF21612 antibody and antibody sequences (listed in FIG. 29 ) in the first two principal components of the sequence space.
  • FIG. 31 illustrates a corresponding plot of all loads describing the influence of each variable on the sample distribution of FIG. 30
  • FIG. 32 provides magnified detail of the bottom center from FIG. 31
  • FIG. 33 provides principal component analysis-derived loads for individual amino acids responsible for clustering of sequences in group containing the sequence 5-a
  • FIG. 2 A general antibody humanization and/or maturation scheme is shown in FIG. 2 . These steps found in FIG. 2 will be briefly introduced here and described in more detail below.
  • Step 01 An antibody or a plurality of antibodies, that partially or fully achieves the desired property (e.g., function, being humanized and/or matured) is used as a starting point (step 01 ).
  • the desired property e.g., function, being humanized and/or matured
  • Step 02 Substitutions to a sequence of step 01 are identified using a combination of changes to the antibody sequence. Such changes are either in monomer identity or in monomer physico-chemical properties. These changes span either the CDR and/or the framework region of heavy chain and/or the light chain of the antibody. For example, consider the case in which the heavy chain of the antibody is being humanized.
  • step 02 a determination can be made that the 21 st and 49 th positions of the heavy chain (based on the kabat numbering scheme) can be changed.
  • a determination is made as to which substitutions can be made at such positions in step 02 . For instance, step 02 may not only determine that the 21 st position of the antibody can be changed, but may also determine that this position should be changed to a glycine, alanine, or leucine.
  • rules are used to determine which positions of the antbodies of step 01 can be changed. Each such rule scores or ranks individual substitutions based on different methods and based on the nature of optimization (i.e) humanization or maturation.
  • Representative rules include, but are not limited to, rules based on (i) changes found in functional, structural or sequence classes, (ii) changes predicted to be favorable using substitution matrices, (iii) changes predicted using evolutionary analysis of the antibody structural and sequence classes, (iv) changes seen in random mutageneis screening, (v) changes predicted by structural modeling, (vi) changes proposed by an expert on the antibody and (vii) changes predicted to be favorable using structural information (vii) changes derived from comparing the framework region of the antibodies with human germline sequences (viii) changes derived from comparing the framework regions of human antibodies (ix) changes derived from substitution matrices constructed from the positional frequencies of amino acids in the CDR regions of all antibodies. (x) Any number of rules can be applied to the one or more antibodies of step 01 .
  • each independent rule assigns a score for each possible substitution position (e.g. residue) in the antibody of step 01 .
  • the scores generated by each of the rules are then combined by methods and/or filters to determine the positions in the antibody that are suitable for change.
  • These scores generated by each of the rules are specific to nature of the optimization process, (i.e) scores are independently derived for humanization of antibodies and for maturation of antibodies.
  • Step 03 identified a set of candidate substitution positions in the antibodies of step 01 .
  • a variant set incorporating such candidate substitutions is designed such that each candidate substitution is tested in combination with many different other candidate substitutions in order to cover the possible search space as evenly as possible (step 03 ).
  • step 01 is a murine antibody and the 2 nd , 5 th , and 15 th kabat positions of the heavy chain has been identified as candidate substitution positions in step 03 .
  • step 02 will constrain the types of amino acids that can be substituted at these positions based on the rules described above. Nevertheless, the full antibody sequence space proposed in step 02 even after filtering can be large.
  • Step 03 seeks to minimize the number of variants that are constructed in order to evenly search and sample this large sequence space.
  • Step 04 Variant antibodies selected in step 03 are individually synthesized and tested for function(s) of interest in step 04 .
  • the variant antibodies are synthesized individually it is easier to keep the number of changes and the number of variants synthesized and tested in each iteration of the process relatively small. In some embodiments, between 5 and 200, more preferably between 10 and 100, and even more preferably between 15 and 50 variants are synthesized and tested in step 04 . By minimizing the number of variants synthesized and tested, relatively inaccurate high throughput assay screens can be avoided in step 04 .
  • Step 05 Various machine-learning methods or other data-mining techniques are used to model the relationship between the sequences and activities of the variant antibodies in step 05 .
  • Step 06 The assessments of the affect of each substitution upon the properties (functions) of the antibodies by each model tested in step 05 are combined in step 06 .
  • Step 07 The assessments of the affect of each substitution upon the properties (functions) of the antibodies by each tested model that was made in step 06 is used in step 07 to design a new set of variant antibodies for synthesis and testing
  • Steps 04 through 07 are repeated a number of times.
  • Each iteration of steps 04 - 07 seeks to design a set of high scoring and diverse antibodies for synthesis and functional testing.
  • Each new set of measurements from an iteration of step 04 is used to refine the sequence-activity model until an end point is reach, at which point the method progresses to step 08 .
  • Step 08 The performance of the methods used to select substitution positions in step 02 and to model the sequence-activity relationships in instances of step 05 are assessed by analyzing the sequences of the best performing variants.
  • the best performing variants are any variants in any iteration of the cycle defined by steps 04 - 07 that score best in one or more functional assays for the target antibody.
  • Step 08 provides a method for tuning the adjustable parameters of the system. Once these parameters have been adjusted, steps 02 through 07 , including multiple iterations of the cycle defined by steps 04 - 07 , are repeated.
  • one of the adjustable parameters of the system is the individual weights for each of the methods applied in step 02 .
  • step 02 method that were good at identifying substitution positions associated with high scoring antibody variants are up-weighted in the next instance of steps 02 through 07 .
  • the modification of weights applied to methods in step 02 based on the results of cycles of steps 04 - 07 allows the system to learn from previous results thereby improving the accuracy with which the system can identify beneficial substitutions (in step 02 ) and assess the contribution of substitutions to antibody activity (in steps 05 and 06 ).
  • FIG. 1 details an exemplary system that supports the functionality described above.
  • the system is preferably a computer system 10 having:
  • a central processing unit 22 a central processing unit 22 ;
  • main non-volatile storage unit 14 for example a hard disk drive, for storing software and data, the storage unit 14 controlled by storage controller 12 ;
  • system memory 36 preferably high speed random-access memory (RAM), for storing system control programs, data, and application programs, comprising programs and data loaded from non-volatile storage unit 14 ; system memory 36 may also include read-only memory (ROM);
  • RAM random-access memory
  • ROM read-only memory
  • a user interface 32 comprising one or more input devices (e.g., keyboard 28 ) and a display 26 or other output device;
  • a network interface card 20 for connecting to any wired or wireless communication network 34 (e.g., a wide area network such as the Internet);
  • any wired or wireless communication network 34 e.g., a wide area network such as the Internet
  • a power source 24 to power the aforementioned elements.
  • Operating system 40 can be stored in system memory 36 .
  • system memory 36 includes:
  • file system 42 for controlling access to the various files and data structures used by the present invention
  • case-specific data 110 and knowledge base 108 each independently comprise any form of data storage system including, but not limited to, a flat file, a relational database (SQL), and an on-line analytical processing (OLAP) database (MDX and/or variants thereof).
  • case-specific data 110 and/or knowledge base 108 is a hierarchical OLAP cube.
  • case-specific data 110 and/or knowledge base 108 comprises a star schema that is not stored as a cube but has dimension tables that define hierarchy.
  • case-specific data 110 and/or knowledge base 108 is respectively a single database.
  • case-specific data 110 and/or knowledge base 108 in fact comprises a plurality of databases that may or may not all be hosted by the same computer 10 .
  • some component databases of case-specific data 110 and/or knowledge base 108 are stored on one or more computer systems that are not illustrated by FIG. 1 but that are addressable by wide area network 34 .
  • FIG. 1 can be located on one or more remote computers.
  • user interface module 104 and other modules can reside on a client computer that is in communication with computer 10 via network 34 .
  • user interface 104 can be an interactive web page.
  • the case-specific data 110 and/or knowledge base 108 and modules (e.g. modules 100 , 104 , 112 , 106 , 116 , 114 , 118 , 130 , 132 ) illustrated in FIG. 1 are on a single computer (computer 10 ) and in other embodiments such data is hosted by several computers (not shown). Any arrangement of case-specific data 110 and knowledge base 108 and the modules illustrated in FIG. 1 on one or more computers is within the scope of the present invention so long as these components are addressable with respect to each other across network 34 or by other electronic means. Thus, the present invention fully encompasses a broad array of computer systems.
  • Expert system 100 is a software module that includes stored knowledge and solves problems in a specific field (for example antibody engineering) by emulating some of the decision processes of a human expert(s).
  • the first set of algorithms that chooses the substitutions and the sequence space to explore for antibody engineering may require expertise in the domains of polynucleotide structure and function, antibody structure and function, protein structural analysis and interpretation, protein structure and function, protein and nucleic acid phylogeny and evolution, chemical and enzymatic mechanisms, bioinformatics and related fields.
  • Expert system 100 applies the knowledge to problems specified by a user who is not necessarily an expert in the domain(s). This invention describes the construction and use of expert system 100 for selecting substitutions useful for mapping and engineering antibody functions.
  • Two functions expert system 100 provides in order to define a sequence space to search are (i) the identification of one or more positions, in the antibody at which substitution is likely to be accepted and where at least some substitutions, insertions, deletions or modifications are likely to result in a functional antibody and (ii) the identification of residues or modifications that are likely to result in a functional antibody when used to substitute or insert at each of the one or more positions identified in (i).
  • An additional or alternative purpose of expert system 100 is the identification of residues or modifications that are likely to affect the desired properties or functions of the antibody. These functions are represented as step 02 in FIG. 2 .
  • One aspect of this invention is the use of methods to identify positions that can be varied, then to synthesize a set of antibody variants containing these substitutions and to test the antibodies for one or more property or function, with the aim of deriving relationships between antibody sequence and function.
  • a user can interact with expert system 100 using user interface 104 .
  • user interface 104 comprises menus, natural language or any other style of interaction.
  • Expert system 100 uses inference engine 106 to reason using the expert knowledge stored in knowledge database 108 together with case-specific data 110 relating to the specific antibody or class of antibodies to be mapped and/or engineered.
  • Case-specific data 110 can be acquired as input from the user of expert system 100 , presented in knowledge base 108 , or acquired from case-specific knowledge generated by the results of experimentation and the analysis facilitated by sequence-activity correlating methods of this invention described in further detail below. These sequence-activity correlating methods are performed in step 05 of FIG. 2 , for example. The data from these sequence-activity correlating methods can additionally be used to add to or alter the information contained within knowledge base 108 .
  • An exemplary rule 120 is:
  • Another exemplary rule 120 is:
  • rules 120 are each of the filters described in FIGS. 4 and 5.
  • Case-specific data 110 can be precompiled by experts. It can also be obtained as user response to questions contained in a component of expert system 100 , for example user interface 104 , knowledge base 108 or inference engine 106 .
  • rules 120 of expert system 100 can also be obtained, in part, by a set of automatic actions executed using one or more computational processes 118 .
  • An example of a computational process 118 is:
  • a target sequence Upon input of a target sequence (from Step 01) ⁇ 202 Search one or more sequence databases for homologs of the target antibody sequence. Store any such sequences in knowledge base 108 204 Identify any functional information provided for any of these target antibody sequences by any of these databases. Store any such functional information in knowledge base 108 206 Search one or more structure databases for homologs of the target antibody sequence. Store any such homolog structural information in knowledge base 108. 208 Search one or more databases for known variants of the target antibody sequence. Store any sequence and functional information in knowledge base 108. 210 Compute the scores for every enumerated substitution found in steps 202 through 208 using select rules 120.
  • Computational processes 118 can be stored in knowledge base 108 as illustrated in FIG. 1 , in expert system 100 , or in any data structure that is accessible by expert system 100 .
  • Some embodiments of expert system 100 include explanation subsystem 112 .
  • Explanation subsystem 112 provides reasons to the user for why particular substitutions are selected by rules 120 .
  • Some embodiments of expert system 100 include knowledge base editor 114 to allow an administrator to add, delete, or modify components of knowledge base 108 including, but not limited to,
  • expert system 100 provides scores for each substitution enumerated along with the contribution to that score from various methods 130 used to evaluate the desirability of each substitution.
  • the weights 132 for the various methods 130 are derived from knowledge base 108 and can be updated by an expert using knowledge base editor 108 and can also be updated automatically using rules in knowledge base 108 .
  • Inference engine 106 is a software module that reasons using information stored in knowledge base 108 .
  • One embodiment of inference engine 106 is a rule-based system. Rule-based systems typically implement forward or backward chaining strategies. Inference engine 106 can be goal driven using backward chaining to test whether some hypothesis is true, or data driven, using forward chaining to draw new conclusions from existing data.
  • Various embodiments of expert system 100 can use either or both strategies.
  • some topics that can be posed by expert system 100 in a goal driven/backward chaining strategy can include: (i) how conservative should an approach be, (ii) how many iterations of the process are likely to achieve the activity of interest, (iii) by what factor should the desired activity increase, and (iv) descriptions of any prior experiments that have failed and why they have failed. Answers to these topics allows expert system 100 to access information from experiments and data from the scientific literature or from personal communications that can be relevant for the design of the sequence space of interest.
  • Inference engine 106 can calculate a probability that a variant residue will provide a desired activity in an antibody of interest.
  • the antibody can be an Fab fragment, (Fab) 2 fragment, a scFv, fragment, a polynucleotide having its own activity of interest, a polynucleotide that encodes an antibody having an activity of interest, or a polynucleotide that encodes a polypeptide that is responsible for synthesis of an antibody having an activity of interest.
  • a profile 116 can be created by inference engine 106 based on probability scores and weighting factors.
  • inference engine 106 calculates the probability that defined substitutions will result in an antibody having the desired function, for any variant of the reference antibody.
  • knowledge base 108 can contain information describing residue positions in the reference sequence that exhibit a high degree of variance in homologs or among sequences in the same structural or sequence class. Inference engine 106 may thus give a high probability that substitutions at such positions will be active.
  • One method of calculating the degree of amino acid variance is described by Gribskov, 1987, Proc Natl Acad Sci USA 84, 4355.
  • a sequence alignment can be available in knowledge base 108 to serve as the basis of a Hidden Markov model that can be used to calculate the probability that one specific residue will be followed by a second specific residue.
  • Hidden Markov models also include probabilities for gaps and insertions. See, Krogh, “An introduction to Hidden Markov models for biological sequences,” in Computational Methods in Molecular Biology , Salzberg et al., eds, Elsevier, Amsterdam. Such models can be used by inference engine 106 to calculate the probability that a particular substitution will possess a desired function.
  • substitution matrix 122 stored in knowledge base 108 can be used by expert system 100 to identify suitable replacement residues for positions likely to accept substitutions.
  • substitutions specific for antibody framework regions and antibody CDR regions can be generated using the sequences in the database.
  • substitutions based on the amino acid frequencies compiled for every CDR position for every antibody class in the kabat database can be derived.
  • the availability of a replacement residue that is likely to be functional can itself determine whether or not a position is likely to accept substitutions. This can be generated from functional sequences that are naturally occurring and/or generated synthetically whose properties have been measured. Substitution matrix 122 choices will impact the probability calculated for likely functionality of a variant.
  • substitution matrix 122 derived from the set of sequences should be chosen.
  • substitution matrix 122 reflecting this need should be chosen.
  • Substitution matrices 122 can be calculated based on the environment of a residue, e.g., inside or accessible, in coil or in beta-sheet. See, for example, Overington et al., 1992, Protein Sci 1:216.
  • Conservation indices 124 stored in knowledge base 108 can be also be used by inference engine 106 to calculate probabilities that a substitution will result in an antibody with desired properties. In this capacity, one can avoid mutating residues that are highly conserved, or conversely, focus mutations on conserved regions of the antibody. Algorithms for calculating conservation indices 124 at each position in a multiple sequence alignment are known in the art. See, for example, Novere et al., 1999 Biophys. Journal 76:2329-2345.
  • Inference engine 106 can also use knowledge of the effects of single mutations as a factor in calculating the probability that a substitution will possess a desired function when mutation effect data 126 is stored in knowledge base 108 .
  • Mutation effect data 126 can originate, for example, from mutagenesis scans or from those substitutions found in naturally occurring variants that affect the function of interest.
  • Inference engine 106 can also use structural information 128 (e.g., crystal structure, insilico models of antibodies, de novo modeled antibody, etc.) stored in knowledge base 108 .
  • structural information 128 e.g., crystal structure, insilico models of antibodies, de novo modeled antibody, etc.
  • inference engine 106 can assign higher probabilities to amino acid residues in framework regions that are close to the CDR of an antibody, as will affect activity and/or specificity than more distant residues.
  • proximity to an epitope, proximity to an area of structural conflict, proximity to a conserved sequence, proximity to a binding site, proximity to a cleft in the protein, proximity to a modification site, etc. can be calculated from structural information 128 and used to calculate the probability that a substitution will result in a functional antibody.
  • the distance of a residue from a region of functional interest physical distances obtained using a known crystal structure of the reference sequence can be used.
  • molecular modeling approaches can be used.
  • the structure of the reference sequence can be predicted based on its homology to a known structure, and then used to calculate distances.
  • the entire structure of the reference sequence can be predicted and distances then calculated from the predicted structure.
  • structural information 128 is energy minimized.
  • the behavior of an antibody can be modeled using molecular dynamic simulations.
  • a crystal structure or a predicted structure can be subjected to molecular dynamic simulation in order to model the effect of various external conditions such as the presence of solvent, the effect of temperature and ionic strength, upon the determined or predicted structure.
  • sequence analysis including sequence complexity, sequence content and composition, internal base-pairing and secondary structure predictions
  • sequence comparisons including structure-based sequence alignments, homology-based sequence alignments, phylogenetic comparisons based on multiple pairwise comparisons, phylogenetic comparisons based on principal component analysis of sequence alignments, Hidden Markov models), evolutionary molecular analysis, structural analysis (including those using X-ray crystallographic data, nuclear magnetic resonance studies, structure threading algorithms, molecular dynamic simulations, active site geometry, determination of surface, internal and active site residues), known or predicted data relating sequence or structure to functional mechanisms, chemical and biophysical properties of functional groups, known or predicted functional effects of changes (for example information derived from the Protein Mutant Resource database,
  • knowledge base 108 is optionally preprocessed for information by knowledge base editor 114 .
  • knowledge base 108 can contain all available antibody sequences.
  • sequences can be, for example, (i) aligned and distributed on a phylogenetic tree, (ii) grouped by principal component analysis (PCA), (iii) grouped by nonlinear component analysis (NLCA) (iv) grouped by independent component analysis (ICA), used to create sequence profiles (see, for example Gribskov, 1987, Proc Natl Acad Sci USA 84, 4355), (v) used to create Hidden Markov models or (vi) used to calculate structures prior to interrogation by the user (vii) classified into canonical structural classes as defined by Chothis and lesk (REF).
  • PCA, NLCA, and ICA is described in, for example, Duda et al., Pattern Classification , Second Edition, John Wiley & Sons, Section 10.13,
  • the output from an expert system 100 will describe the various substitutions recommended by methods 130 based on assignment of scores, confidences, ranks, or probabilities (hereinafter “scores”) using rules 120 in knowledge base 108 .
  • these scores are cumulative. That is, every rule 120 used by a method 130 will assign a score to the substitution under consideration and these scores can be higher if more rules are satisfied.
  • FIG. 3 shows a series of steps that can be executed by expert system 100 in order to identify substitutions that are likely to increase the ability of an antibody to bind to a specific target antigen.
  • Five independent methods 130 are shown for assessing the suitability of a substitution in the framework and CDRs: (i) substitutions from antibody sequences derived from other species and/or from synthetically derived antibodies and/or germline sequences from human and/or other species (ii) substitutions from homologous and modeled structures, (iii) substitutions from substitution matrices, (iv) substitutions from principal component analysis (PCA) and (v) substitutions from binding pocket analysis.
  • PCA principal component analysis
  • substitutions from binding pocket analysis For each method 130 , one or more rules (filters) 120 defined in knowledge base 108 are used.
  • substitutions from homologous structures uses two rules 120 .
  • the first rule 120 is an estimate of the mean root mean square deviation (RMSD) from the target structure for every five residue window of the homolog structure, and select framework sites that deviate from the target structure by more than three A.
  • the second rule 120 identifies amino acid substitutions that are found in homologous sequences and select framework sites that are within five A of the complementarity determining region.
  • rules 120 are applied as filters: a substitution that satisfies one of the rules is considered to have passed through that filter and receives a score. For example, in FIG. 3 , this score is 1.
  • step 02 of FIG. 2 uses the following algorithm in order to identify suitable substitutions:
  • Rules 120 can also be combined on a case by case basis, using expert knowledge. These rules 120 can be stored in a knowledge base 108 and can be executed by inference engine 106 using user input acquired by questioning the user for requirements and knowledge via the user interface 104 .
  • each rule 120 produces a reproducible quantitative value that can be used as a measure of the suitability of a substitution.
  • a rule 120 can be used to produce an absolute quantitative score. This absolute quantitative score can be used directly, or it can be used to create a rank order list or a filter. As an example consider rule 1 b of FIG. 3 . Rule 1 b calculates the difference in free energy between a target antibody and an antibody containing a substitution. This value can then be used in several different ways to compare the favorability of different substitutions.
  • the absolute value of the free energy difference (caused by the substitution) can be used, (ii) the free energy differences of all possible substitutions can be ranked in order of favorability, then a subset of substitutions that are predicted to be the most favorable can be selected and assigned a score, (iii) the score can be a single value assigned to all of the substitutions belonging to the subset of the most favorable, (iv) the score can be a measure of the rank order of the substitution, so that the most favorable substitutions receive a higher score than those that are calculated to be less favorable, (v) a rule can also be used to rank all possible substitutions in order of predicted favorability and then eliminate a subset of these substitutions that are predicted to be the least favorable. In option (v), substitutions that were eliminated would receive a score of zero.
  • the scores produced by individual rules can be combined in a variety of ways. In some embodiments they are added together in the manner illustrated in the algorithm illustrated in Section 5.1 above. In some embodiments, the scores are multiplied together. For example,
  • one or more rules 120 can be used as a filter, so that only substitutions passing the one or more filter are used, regardless of their scores from the other rules. For example,
  • a cumulative score can be produced by any mathematical function of the scores produced by two or more individual rules. For example,
  • weight is some rule 120 specific weight that is independently assigned to a rule.
  • Such weights can be stored in knowledge base 108 and adjusted by an expert using knowledge-base editor 114 ( FIG. 1 ).
  • scores produced by individual rules can be scaled or normalized and/or transformed by a mathematical function to facilitate their combination.
  • the mutation I46V was identified as the most favorable substitution in the framework region by combining the scores from methods 130 .
  • the distance, expressed in fraction of amino acid differences, was transformed and a Poisson correction ( ⁇ log [1-fraction]) applied and multiplied by the product of the absolute scores obtained from the other methods 130 .
  • the resulting scores for all substitutions were ranked and I46V (combination score 126 ) was ranked 1.
  • different criteria were used to compute the scores for the framework and CDR regions.
  • the scores produced by individual rules 120 can be assigned different weights prior to being combined. For example, if the total score for a substituting monomer x at position i (S ix ) is obtained by adding the scores obtained by applying n different rules, the score can be expressed by Equations (1) or (2):
  • Rules (and weights) can be (i) specific for a substitution of monomer x at a specific location, (ii) specific for position for any and/or a group of monomer substitution(s), (iii) specific for any and/or a group of positions for a specific monomer x, (iv) specific for any substitutions derived from a particular and/or a group of homologs, (v) or specific for any position derived from a particular and/or a group of homologs.
  • weights to modify scores obtained using different rules 120 allows different rules 120 to have different degrees of influence over the final score for a substitution. For example if Rule 4 is the most important in determining the suitability of a substitution in a particular antibody, then this rule can be made to dominate the total score for the substitutions by making W 4 much higher than the other weights.
  • weights to modify scores obtained using different rules 120 allows different rules 120 to have different degrees of influence over the final score for a substitution depending upon the class or subclass of antibodies being considered. For example a rule 120 considering the structural effect of a substitution can be most important for engineering an antibody, while a rule 120 considering the statistical likelihood of a substitution using a substitution matrix can be most important for engineering a protease.
  • expert system 100 can then be used to assign weights to the scores from different rules 120 that will result in the most accurate assessment of the favorability of substitutions.
  • expert system 100 can assign different weights to different methods, to produce more control over how substitutions scores are computed.
  • weights to modify scores obtained using different rules 120 allows expert system 100 to incorporate information obtained from previous experiments.
  • another aspect of the invention involves the use of sequence-activity relationships to empirically measure the contribution of substitutions to one or more activity of an antibody. This aspect of the invention is described more fully in Section 5.5.
  • This sequence-activity determination effectively creates a feedback loop by which weights assigned to the scores from different rules 120 applied by expert system 100 can be adjusted.
  • 20 substitutions within an antibody represented by S 1 —S 20
  • a set of antibodies that contain these substitutions are synthesized, and a sequence-activity relationships derived using wet lab assays.
  • the sequence-activity relationships are used to determine actual scores that measure the fitness of each substitution for the desired activity of the antibody (F 1 -F 20 ).
  • the weights applied to each rule 120 and/or method 130 can then be adjusted so that the observed fitness of each substitution, F 1 -F 20 , correlate more closely with scores C 1 -C 20 produced by expert system 100 .
  • this correlation is the correlation between the absolute values of the scores for each substitution from expert system and the observed fitness of each substitution derived from the sequence-activity relationship.
  • the correlation can be a correlation between the rank order of effect of substitutions predicted by expert system 100 and the rank order of substitutions observed or derived from the sequence-activity relationship.
  • the weights applied to each rule 120 can also be adjusted so that the correlation between the observed fitness of substitutions and the scores produced by expert system 100 is maximized for more than one set of substitutions, in one or more different target antibodies.
  • Different classes of antibodies can optionally be used to provide different sets of substitutions for comparing observed fitness and scores produced by expert system 100 . This allows different weights to be calculated to apply to the scores produced by different rules 120 as a function of antibody class.
  • One skilled in the art will appreciate that there are many possible variations of using experimental results to adjust weights applied to rule 120 scores. All such variants, whose predictive scoring functions can be adjusted based upon experimental data, are within the scope of the expert systems 100 of the present invention and can thus be considered systems capable of learning.
  • the score for a substitution based on two or more rules can be calculated independently or using conditional probabilities.
  • An expert system 100 can produce scores for at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45, 50, 60, 70, 80, 90 or 100 positions in the reference sequence up to the entire sequence, and can include contiguous residues or noncontiguous residues or mixtures thereof.
  • the expert system 100 can include at least 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 25, 30, 35, 40, 45 or 50 different residues.
  • Naturally occurring residues can be included in the expert system, as well as unnatural residues for synthetic methods, and combinations thereof.
  • the above calculations can be performed by an expert with access to the relevant knowledge base 108 , for example, by using user interface 104 .
  • FIGS. 1 and 2 Examples of the ways in which such expert system 100 can be used to automatically select substitutions to make in an antibody will now be described in the following sections with reference to FIGS. 1 and 2 .
  • the following exemplary process is intended to illustrate one possible embodiment of the invention.
  • One skilled in the art will recognize that there are many possible variations on this theme, and the following is not intended to limit the present invention.
  • the selection process refers to the scheme shown in FIG. 3 .
  • FIG. 3 shows a series of independent rules 120 , each of which can be used to produce a score for any possible amino acid substitution in an antibody.
  • all possible single substitutions can be enumerated computationally and then scored according to one or more of the rules executed by expert system 100 .
  • One source of information that can be used to construct rules 120 that assess the likely effect of amino acid substitutions upon one or more activities of an antibody is the sequence of one or more homologous or related antibodies. See, for example, FIG. 3 , rule 3 a .
  • Homologous sequences are generally analogous functionally and structurally, although having been subjected separately to different selective pressures they are also likely to be optimized differently.
  • Antibody sequences variants can also be generated in the lab using many techniques and sequence, properties of several such antibodies are available in the database and literature. Amino acids that differ between homologous sequences thus provide a guide to substitutions that are likely to yield functional though different antibody sequences.
  • homologous antibody sequences or sequence classes are aligned (e.g., by using clustalw; Thompson et al., 1994, Nucleic Acids Res 22: 4673-80) and then a phylogenetic tree is reconstructed.
  • Conservation indices can then be calculated for each site (e.g., Dopazo, 1997, Comput Appl Biosci 13: 313-7) and the information content calculated for each site (e.g., Zhang, 2002, J Comput Biol 9: 487-503).
  • These scores can be exhaustively calculated for every position in the antibody. The scores reflect the extent of tolerance to substitutions in the antibody at each position.
  • Scores for a given alignment can also be normalized to have an average value of 0.0 and a standard deviation of 1.0, or other standard procedures can be used to compare and combine scores from multiple methods. These values can then be used directly as a score, as outlined above and in Equation (1) or Equation (2). In some embodiments, all sites with a score above a certain threshold value can be selected. For example, a cutoff (threshold) of 0.0 can be chosen (which is set to be the average score).
  • the most variable (e.g., least conserved) sites can be selected by ranking the sites in order of these scores. For example the most highly scoring site can be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 50, 60, 70, 80, 90 or 100 most highly scoring sites can be selected. In some embodiments the least variable (e.g., most conserved) sites can be eliminated by ranking the sites in order of these scores.
  • the least highly scoring site can be eliminated, or the 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200, 210, 220, 230, 240, 250, 260, 270, 280, 290, 300, 310, 320, 330, 340, 350, 360, 370, 380, 390, 400, 500, 600, 700, 800, 900 or 1000 least highly scoring sites can be eliminated ( FIG. 3 , Rule 1 a ).
  • Amino acid diversity and tolerance at each site can be measured as a fitness property of each amino acid at every location. In this approach we all related antibody sequences available can be considered. The most fit residue for that position carries a higher value (e.g., Koshi et al., 2001, Pac Symp Biocomput 191-202; O. Soyer, M. W. Dimmic, R. R. Neubig, and R. A. Goldstein; Pacific Symposium on Biocomputing 7:625-636 (2002). Sites can be grouped into site-classes or treated independently. Sites and site classes most fit to change based on the substitution rate and the substitutions most favorable based on the fitness can be selected ( FIG. 3 , Rule 2 a ).
  • these values of fitness can then be used directly as a score, as outlined above and in Equation (1) or Equation (2).
  • all sites with a score above a certain threshold value can be selected. For example, a cutoff (threshold) of 0.0 can be chosen (when the normalization of scores sets the wild type residue found in the reference to be 0.0.
  • all sites with a score below a certain threshold value can be eliminated. Threshold values of 0.0 or below can be eliminated, thereby only including amino changes that have a higher fitness value that the reference wild type amino acid found in that position.
  • the sites most tolerant to change can be selected by ranking the sites in order of these scores.
  • the most highly scoring site can be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 50, 60, 70, 80, 90 or 100 most highly scoring sites may be selected.
  • the sites least tolerant to change can be eliminated by ranking the sites in order of these scores.
  • the least highly scoring site can be eliminated, or the 10, 20, 30, 40, 50, 60, 70, 80, 90, 100, 110, 120, 130, 140, 150, 160, 170, 180, 190, 200, 210, 220, 230, 240, 250, 260, 270, 280, 290, 300, 310, 320, 330, 340, 350, 360, 370, 380, 390, 400, 500, 600, 700, 800, 900 or 1000 least highly scoring sites can be eliminated.
  • GPCR G-protein coupled receptors
  • Scores can also be assigned to residues from related sequences that are classified into the same canonical class as the target antibody. In this approach substitutions that are derived from sequences that are part of the same Chothia-Lesk canonical class can be scored ( FIG. 3 Rule 3 a ).
  • candidate amino acid changes can modeled into the structure(s) computationally and changes in the free energy computed. These computationally calculated changes in free energies resulting from the substitutions can then be used directly as a score, as outlined above and in Equation (1) or Equation (2). Alternatively, all changes can be selected that increase the free energy of the antibody by less than a certain value.
  • all changes that would increase the free energy by less than 1 kCal/mol can be selected, all changes that would increase the free energy by less than 1.5 kCal/mol, can be selected, all changes that would increase the free energy by less than 2 kCal/mol can be selected, or all changes that would increase the free energy by less than 2.5 kCal/mol can be selected.
  • all changes can be eliminated that increase the free energy of the antibody by more than a certain value.
  • the best tolerated substitutions can be selected by ranking the sites in order of the predicted increase in free energy.
  • substitution with the lowest increase in free energy can be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 50, 60, 70, 80, 90 or 100 substitution with the lowest increase in free energy may be selected.
  • the substitutions with the greatest increases in free energy can be eliminated by ranking the sites in order of these scores.
  • multiple changes can be modeled into the structure(s) computationally and changes in the free energies resulting from the substitutions computed. These free energy values can be used to identify changes that are “valid” independently, but not together. Amino acid changes that are independent can be selected preferentially. Amino acid clashes that yield a higher free energy when compared to the free energies produced by modeling changes separately can be eliminated.
  • Regions of the antibody that differ structurally between antibodies are more likely to tolerate change, while those regions that are structurally conserved are likely to be less tolerant.
  • Structures can be directly obtained from the database or predicted using various structure modeling software packages. Structures of homologs and mutants can be superposed on the wild type structure. See, for example, May et al., 1994, Protein Eng 7: 475-85; and Ochagavia et al., 2002, Bioinformatics 18: 637-40). Structural conservation can be calculated as the root mean squared (RMS) deviations of the backbones of the superposed chains.
  • RMS root mean squared
  • RMS deviations can be computed as the deviations of individual residues, or more preferably as the deviations of a running average over a between two and ten residue stretch of the backbone between the target antibody and one or more homologous antibodies.
  • These computationally calculated RMS deviations for every position between homologous structures can then be used directly as a score, as outlined above and in Equation (1) or Equation (2).
  • RMS deviations between the alpha carbons (or backbone atoms) in the structure of the target antibody and one or more homologous or related antibodies that are greater than a threshold value can be considered structurally labile and these sites can be selected.
  • This threshold RMS deviation between homologous structures can be greater than 2 ⁇ , 2.5 ⁇ , 3 ⁇ , 3.5 ⁇ , 4 ⁇ , 4.5 ⁇ , 5 ⁇ .
  • RMS deviations between the alpha carbons in the structure of the target antibody and one or more homologous or related antibodies that are less than a threshold value can be considered structurally conserved and these sites can be eliminated.
  • This threshold RMS deviation between homologous structures can be less than 2 ⁇ , 2.5 ⁇ , 3 ⁇ , 3.5 ⁇ , 4 ⁇ , 4.5 ⁇ , or 5 ⁇ .
  • sites can be ranked in order of the calculated RMS deviations between the alpha carbons in the structure of the target antibody and one or more homologous or related antibodies and those with the highest calculated RMS deviations selected.
  • the site with the highest calculated RMS deviations between homologous structure can be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 50, 60, 70, 80, 90 or 100 sites with the highest calculated RMS deviations between homologous structure may be selected.
  • sites can be ranked in order of the calculated RMS deviations between the alpha carbons in the structure of the target antibody and one or more homologous or related antibodies and those with the lowest calculated RMS deviations eliminated.
  • the site with the lowest calculated RMS deviations between homologous structures can be eliminated or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 50, 60, 70, 80, 90 or 100 sites with the lowest calculated RMS deviations between homologous structure can be eliminated ( FIG. 3 , Rule 2 b ).
  • Changes near binding sites (and CDRs) are highly likely to influence the activity of the antibody and are good candidates for substitution.
  • All amino acid substitutions that are found in one or more variants can be tested for proximity to a binding or regulatory site of the antibody.
  • the distance between an amino acid substitution that is found in one or more homologs from a binding or catalytic or regulatory site can be used directly as a score, as outlined above and in Equation (1) or Equation (2).
  • all amino acid substitutions that are found in one or more homologs and that are within a threshold distance of a binding or regulatory site in the antibody can be selected.
  • This threshold distance can be less than 2 ⁇ , 2.5 ⁇ , 3 ⁇ , 3.5 ⁇ , 4 ⁇ , 4.5 ⁇ , 5 ⁇ , 5.5 ⁇ , 6 ⁇ , 6.5 ⁇ , 7 ⁇ , In still other embodiments, all amino acid substitutions that are found in one or more homologs and that are beyond a threshold distance of a binding or regulatory site in the antibody can be eliminated.
  • This threshold distance can be more than 2 ⁇ , 2.5 ⁇ , 3 ⁇ , 3.5 ⁇ , 4 ⁇ , 4.5 ⁇ , 5 ⁇ , 5.5 ⁇ , 6 ⁇ , 6.5 ⁇ , or 7 ⁇ .
  • all amino acid substitutions that are found in one or more homologs can be ranked in order of proximity to a binding or regulatory site in the protein and those that are closest to the binding or regulatory site selected by a rule 120 .
  • the substitution closest to the binding or catalytic or regulatory site can be selected, or between 2 and 20, between 10 and 100, or the top 200 substitutions closest to the binding or catalytic or regulatory site can be selected.
  • all amino acid substitutions that are found in one or more homologs can be ranked in order of proximity to a binding or regulatory site in the antibody and those that are farthest from the binding or catalytic or regulatory site eliminated.
  • the substitution farthest from the binding or regulatory site can be eliminated.
  • between 2 and 20, between 10 and 100, or the top 200 substitutions farthest from the binding or regulatory site can be eliminated.
  • Another source of information that can be used to construct rules 120 that assess the likely effect of amino acid substitutions upon one or more activities of an antibody is the frequency with which one amino acid is observed to substitute for another amino acid in different proteins.
  • the matrix can be expressed in terms of probabilities or values derived from probabilities by mathematical transformation involving probabilities of transitions or substitutions (Pij) and observed frequencies of amino acids (Fi). Matrices using such transformation include scoring matrices like PAM100, PAM250, and BLOSUUM etc. See, for example, FIG. 3 , rule Ic.
  • Substitution matrices are derived from pairwise alignments of protein homologs from sequence databases. They constitute estimates of the probability that one amino acid will be changed to another while conserving function.
  • substitution matrices are calculated from different sets of sequences. For example, they can be based on the structural environment of a residue (Overington, 1992, Genet Eng (N Y) 14: 231-49.; and Overington et al., 1992, Protein Sci 1: 216-26.) or on additional factors including secondary structure, solvent accessibility, and residue chemistry (Luthy et al., 1992, Nature 356: 83-5.
  • Substitution matrices can be derived for specific sites or group of sites in the antibody. Specifically, substitutions specific for antibody framework regions and antibody CDR regions can be generated using the sequences in the database. Additionally, substitutions can be derived based on the amino acid frequencies compiled for every CDR position for every antibody class in the kabat database.
  • a substitution matrix that best captures the observed sequences in the antibody family of interest can be calculated using the Bayesian method developed by Goldstein et al. (Koshi et al., 1995, Protein Eng 8: 641-645) and used to score all candidate substitutions.
  • these values can then be used directly as a score, as outlined above and in Equation (1) or Equation (2).
  • the scores can expressed as Pij: the probability of substituting residue i with j. Any transformations of Pij can also be used. Pij can be computed for a specified evolutionary distance. In alternative embodiments, all substitutions with a probability above a certain threshold value may be selected. Threshold values of 0.00001, 0.00001, 0.0001, 0.01 or 0.1 can be used for probabilities and/or threshold values of ⁇ 5, ⁇ 4, ⁇ 3, ⁇ 2, ⁇ 1, 0, 1, 2, 3, 4, 5 for any PAM matrix. In still other embodiments, all substitutions with a probability below a certain threshold value may be eliminated.
  • Threshold values of 0.00001, 0.00001, 0.0001, 0.01 or 0.1 can be used for probabilities and/or threshold values of ⁇ 5, ⁇ 4, ⁇ 3, ⁇ 2, ⁇ 1, 0, 1, 2, 3, 4, 5 for any PAM matrix
  • the most favorable substitutions can be selected by ranking substitutions in order of their substitution matrix probability scores.
  • the most highly scoring substitution can be selected, or the top 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, up to 50, up to 60, up to 70, up to 80, up to 90, up to 100, up to 110, up to 120, up to 130, up to 140, up to 150, up to 160, up to 170, up to 180, up to 190, up to 200, up to 210, up to 220, up to 230, up to 240, up to 250, up to 260, up to 270, up to 280, up to 290, up to 300, up to 310, up to 320, up to 330, up to 340, up to 350, up to 360, up to 370, up to 380, up to 390, up to 400, up to 500, up to 600, up to 700, up to 800, up to 900, up to 1000, up to 2000, up to
  • the least favorable substitutions can be eliminated by ranking substitutions in order of their substitution matrix probability scores.
  • the least substitution with the lowest substitution matrix probability may be eliminated, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, up to 50, up to 60, up to 70, up to 80, up to 90, up to 100, up to 110, up to 120, up to 130, up to 140, up to 150, up to 160, up to 170, up to 180, up to 190, up to 200, up to 210, up to 220, up to 230, up to 240, up to 250, up to 260, up to 270, up to 280, up to 290, up to 300, up to 310, up to 320, up to 330, up to 340, up to 350, up to 360, up to 370, up to 380, up to 390, up to 400, up to
  • a substitution or a scoring matrix can be calculated by considering homologous and/or related antibodies from many different antibody classes (e.g., Benner et al., 1994, Protein Eng 7: 1323-1332; and Tomii et al., 1996, Protein Eng 9: 27-36) can be used to score all candidate substitutions. In some embodiments, these values can then be used directly as a score, as outlined above and in Equation (1) or Equation (2). In some embodiments, all substitutions with a probability above a certain threshold value can be selected.
  • Threshold values of 0.00001, 0.00001, 0.0001, 0.01 or 0.1 can be used for probabilities and/or threshold values of ⁇ 5, ⁇ 4, ⁇ 3, ⁇ 2, ⁇ 1, 0, 1, 2, 3, 4, 5 for any PAM matrix can be used. In still other embodiments, all substitutions with a probability below a certain threshold value can be eliminated. Threshold values of 0.00001, 0.00001, 0.0001, 0.01 or 0.1 can be used for probabilities and/or threshold values of ⁇ 5, 4, ⁇ 3, ⁇ 2, ⁇ 1, 0, 1, 2, 3, 4, 5 can be used for any PAM matrix. In still other embodiments, the most favorable substitutions can be selected by ranking substitutions in order of their substitution matrix probability scores.
  • the most highly scoring substitution may be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, up to 50, up to 60, up to 70, up to 80, up to 90, up to 100, up to 110, up to 120, up to 130, up to 140; up to 150, up to 160, up to 170, up to 180, up to 190, up to 200, up to 210, up to 220, up to 230, up to 240, up to 250, up to 260, up to 270, up to 280, up to 290, up to 300, up to 310, up to 320, up to 330, up to 340, up to 350, up to 360, up to 370, up to 380, up to 390, up to 400, up to 500, up to 600, up to 700, up to 800, up to 900, up to 1000, up to 2000, up to 3000
  • the least favorable substitutions can be eliminated by ranking substitutions in order of their substitution matrix probability scores.
  • the least substitution with the lowest substitution matrix probability may be eliminated, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, up to 50, up to 60, up to 70, up to 80, up to 90, up to 100, up to 110, up to 120, up to 130, up to 140, up to 150, up to 160, up to 170, up to 180, up to 190, up to 200, up to 210, up to 220, up to 230, up to 240, up to 250, up to 260, up to 270, up to 280, up to 290, up to 300, up to 310, up to 320, up to 330, up to 340, up to 350, up to 360, up to 370, up to 380, up to 390, up to 400, up to
  • Antibody sequences can be mathematically represented in terms many variables, each variable representing the type of amino acid at a specific location.
  • the sequence AGWRY can be represented by 5 variables, where variable 1 assumes a value of “A” corresponding to position 1, variable 2 is “G” corresponding to position 2 and so on.
  • Each variable can assume 1 of 20 possibilities.
  • each position can assume a value corresponding to a physico-chemical property of the amino acid instead of amino acid identity.
  • each variable can be a combination of variables representing properties of amino acids.
  • each variable can be represented in a binary form corresponding to presence or absence of a particular amino acid.
  • each variable can be represented in a binary form corresponding to presence or absence of a defined group of amino acids.
  • Typical antibodies contain many hundred variables.
  • a set of antibodies are various points in the variables space, and relationships between various antibodies can be represented in terms of the values of the variables corresponding to those antibodies.
  • antibodies can be clustered and classified using statistical techniques like the principal components analysis, k-means clustering, SVM etc.
  • PCA Principal Component Analysis
  • Antibody sequence can be represented in terms of the principal components of that sequence. Principal components can then be identified in which antibodies are grouped functionally. The loads of those principal components can then be used to identify the monomers that are most responsible for the grouping of the antibodies within sequence space. These monomers are thus good candidates for substitutions likely to affect function.
  • amino acid substitutions that are most important in differentiating and grouping sequences are often also those that functionally differentiate the proteins.
  • Identification of such amino acids using dimension-reducing techniques such as principal component analysis has been described (e.g., Casari et al., 1995 , Nat Struct Biol 2: 171-178; Gogos et al., 2000, Proteins 40: 98-105; and del Sol Mesa et al., 2003, J Mol Biol 326: 1289-1302).
  • PCA can identify sequence features and substitutions corresponding to the desired phenotype of the protein and scores “loads” for these features in the direction of desired phenotype are used as absolute scores or as filters to identify substitutions.
  • FIG. 8 shows the accession number of the list of 49 proteases whose sequences are homologous to proteinase K.
  • a property of interest in this example is activity during or after exposure of the protein to heat.
  • the 49 sequences were subjected to principal component analysis, and the distribution of the sequences in the first two principal components is shown in FIG. 9 .
  • Proteases 46 , 47 , 48 and 49 were all obtained from thermostable organisms and can thus be expected to possess desirable thermostability properties.
  • these four proteases are grouped together in the first two principal components of the sequence space, characterized by strongly negative scores in both principal components 1 and 2 .
  • FIG. 9 shows the accession number of the list of 49 proteases whose sequences are homologous to proteinase K.
  • a property of interest in this example is activity during or after exposure of the protein to heat.
  • the 49 sequences were subjected to principal component analysis, and the distribution of the sequences in the first two principal components. 9 .
  • FIG. 10 shows the contributions (the “loads”) of all amino acid differences within the alignment of the 49 proteases, to the new dimensions principal components 1 and 2 .
  • FIG. 11 shows an expanded detail of the lower left corner of FIG. 10 in which the identities of each amino acid contributing to the principal components are now shown. These amino acids are those most responsible for giving a protein sequence a strong negative score in principal component 1 and principal component 2 . These contributions are quantitated in FIG. 12 . Because these scores are also those seen for proteases from thermophilic organisms, the amino acids that are primarily responsible for conferring these scores upon proteins are very good candidates for amino acids that may confer desirable properties, in this case thermostability.
  • FIGS. 24-28 An example of the use of principal component analysis related to antibody humanization for identification of favorable substitutions is also shown in FIGS. 24-28 .
  • FIG. 24 shows the sequence identification number listed by locus of germline sequence from VBase (available at http://www.mrc-cpe.cam.ac.uk/).
  • a properties of interest in this example are characteristics of sequence 4-28.
  • the heavy chain of sequences listed in FIG. 24 along with the heavy chain of the murine antibody RSV19 were subjected to principal component analysis, and the distribution of the sequences in the first two principal components is shown in FIG. 25 .
  • Sequence 4-28 is the sequence cluster containing sequences close in locus id. As shown in FIG. 25 , these are grouped together in the first principal components of the sequence space, characterized by strongly positive scores in principal component 1 .
  • FIG. 25 shows the sequence identification number listed by locus of germline sequence from VBase (available at http://www.mrc-cpe.cam.ac.uk/).
  • a properties of interest in this example are characteristics of sequence 4-28.
  • FIG. 26 shows the contributions (the “loads”) of all amino acid differences within the alignment, to the new dimensions principal components 1 and 2 .
  • FIG. 27 shows an expanded detail of the right center of FIG. 26 in which the identities of each amino acid contributing to the principal components are now shown. These amino acids are those most responsible for giving a protein sequence a strong positive score in principal component 1 . Some of these contributions are quantitated in FIG. 28 . The amino acids that are primarily responsible for conferring these scores upon proteins are serve as candidates for amino acids that may confer desirable properties, in this case characteristics of germline sequence 4-28.
  • FIG. 29 shows the sequence identification number listed by locus of germline sequence from VBase (available at http://www.mrc-cpe.cam.ac.uk/).
  • a property of interest in this example are characteristics of sequence 5-a.
  • the heavy chains of germline sequences along with the heavy chain of AAF21612 were subjected to principal component analysis, and the distribution of the sequences in the first two principal components is shown in FIG. 30 .
  • Sequence 5-a is in the sequence cluster containing sequences close to locus id 1-x. As shown in FIG.
  • FIG. 31 shows the contributions (the “loads”) of all amino acid differences within the alignment, to the new dimensions principal components 1 and 2 .
  • FIG. 32 shows an expanded detail of the lower center of FIG. 31 in which the identities of each amino acid contributing to the principal components are now shown. These amino acids are those most responsible for giving a protein sequence a strong negative score in principal component 2 . Some of these contributions are quantitated in FIG. 33 .
  • the amino acids that are primarily responsible for conferring these scores upon proteins are very good candidates for amino acids that may confer desirable properties, in this case characteristics of germline sequence 5-a.
  • any sequence principal component can be used that contributes to differentiating between two sets of antibodies and that is likely to reflect some functional differences of interest.
  • the “load” contributed by a substitution to one or more such principal component of sequence can be used directly as a score, as outlined above and in Equation (1) or Equation (2).
  • all substitutions with a “load” above a certain threshold value can be selected. Threshold values can be determined from the distribution of load values. For example, select top ten percent positive loads in principal component 1 .
  • all substitutions with a “load” below a certain threshold value can be eliminated. For example, eliminate the top ten percent of the negative loads in principal component 1 .
  • the substitutions with the highest loads can be selected by ranking substitutions in order of their loads.
  • the substitution with the highest “load” can be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, 50, 60, 70, 80, 90 or 100 substitutions with the highest “loads” can be selected.
  • the substitutions with the lowest loads can be eliminated by ranking substitutions in order of their loads.
  • the substitution with the lowest “load” can be eliminated, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19, 20, 21, 22, 23, 24, 25, 26, 27, 28, 29, 30, 31, 32, 33, 34, 35, 36, 37, 38, 39, 40, up to 50, up to 60, up to 70, up to 80, up to 90, up to 100, up to 110, up to 120, up to 130, up to 140, up to 150, up to 160, up to 170, up to 180, up to 190, up to 200, up to 210, up to 220, up to 230, up to 240, up to 250, up to 260, up to 270, up to 280, up to 290, up to 300, up to 310, up to 320, up to 330, up to 340, up to 350, up to 360, up to 370, up to 380, up to 390, up to 400, up to 500, up to 600, up to 700, up to 800, up to 900, up to 1000, up to 2000,
  • a set of substitutions can be identified for testing. These may be the substitutions with the highest aggregate scores, they may be the substitutions with the highest score for each individual rule 120 , or they may be derived in some other way using the scores produced by the rules 120 used by methods 130 of expert system 100 .
  • the number of substitutions selected by step 03 of FIG. 2 in one cycle of the optimization process is less than 1000 substitutions, more preferably less than 250 substitutions, more preferably less than 100 substitutions and more preferably less than 50 substitutions.
  • the rules discussed in Section 5.1 above and shown in FIG. 3 are one example of the way in which an initial sequence space can be defined.
  • the sequence space is defined in terms of an initial target antibody sequence, and substitutions to be made in that target sequence. Each substitution is defined in terms of a position in the target antibody, and the identity of a monomer with which the monomer at that position in the target antibody is to be replaced. Selection of the target antibody corresponds to step 01 in FIG. 2 . Definition of the sequence space corresponds to step 02 in FIG. 2 . This section is directed to step 03 of FIG. 2 .
  • this designed antibody variant set includes only a subset of the total number of possible variants that could be generated.
  • the total number of possible variant proteins in a sequence space defined by a target antibody containing all possible combinations of 24 substitutions is 2 24 >16,000,000.
  • the methods of the present invention allow the interrogation of this sequence space by designing and synthesizing only a very small fraction of the total number of antibodies that are included in the sequence space defined by the initial target antibody and the substitutions.
  • the number of variants in the designed antibody variant set is less than 1000 variants, more preferably less than 250 variants and more preferably less than 100 variants. This is possible because, although the designed antibody variant set includes only a subset of the total number of possible variants (e.g. the possible combinations of substitutions), care is taken to test all antibody substitutions in many different sequence contexts.
  • An example is shown in FIG. 14 , where a set of 24 variants were designed to interrogate the sequence space defined by a target antibody sequence and 24 substitutions in FIG. 13 .
  • each variant contains six substitutions, each substitution occurs six times within the designed antibody variant set, and each occurrence of each substitution takes place within a quite different context, that is it is combined with a different set of other substitutions each time.
  • the aim when designing a set of antibody variants to interrogate a sequence space defined by a target antibody sequence and a set of substitutions is to obtain a designed antibody variant set where the substitutions are distributed in such a way that a large amount of information can subsequently be extracted from sequence-activity relationships.
  • the design of antibody variant sets has common elements with the design of experimental datasets from a diverse range of other disciplines including agriculture and engineering. Methods to optimize experimental datasets (experimental design or design of experiment: DOE) are described by Sir R. A. Fisher in 1920 (Fisher, The Design of Experiments, MacMillan Publishing Company; 9th edition, 1971).
  • Plackett and Burman developed the idea further with the introduction of screening designs (e.g., Plackett et al., 1946, Biometrika 33: 305-325), and Taguchi subsequently introduced the orthogonal matrix (Taguchi, 1986 , Introduction to Quality Engineering , Asian Productivity Organization, Distributed by American Supplier Institute Inc., Dearborn, Minn.).
  • Any number of experimental design techniques can be used to maximize the information content of the designed antibody variant set including, but not limited to, complete factorial design, 2 k factorial design, 2 k fractional factorial design, central composite, latin squares, Inc., Inc., Dearborn, Minn.).
  • Any number of experimental design techniques can be used to maximize the information content of the designed antibody variant set including, but not limited to, complete factorial design, 2 k factorial design, 2 k fractional factorial design, central composite, latin squares, Inc.-Burmann designs, Taguchi design, and combinations thereof.
  • the methods described above were designed to maximize the amount of information that could be obtained from a specified limited number of experiments that could be performed. This is conceptually comparable to the resource limitation seen in antibody optimization, where functional tests are complex and time, cost or other resource-limited. However, a significant difference between antibody optimization and other applications of experimental design is that for antibody optimization there is an additional constraint.
  • the simultaneous introduction of many changes can adversely affect functional properties of the antibody.
  • a variant references to an antibody that has a sequence that is identical to the sequence of the antibody selected in step 01 of FIG. 2 with the exception that there are one or more substitutions in the sequence.
  • a substitution refers to a mutation at a particular position in the antibody from the residue found at that position in the antibody selected in step 01 of FIG. 2 to some other residue.
  • any method can be appropriate provided that the number of substitutions in each variant set is relatively small so that the majority of antibodies are active.
  • the number of previously untested substitutions present in each variant is preferably 9, 8, 7, 6, 5, 4, 3 or 2.
  • each selected substitution be tried an approximately equal number of times in the designed antibody variant set.
  • each substitution be tested in many different sequence contexts. In other words each substitution appears in a number of different antibody variants, in each case being combined with a different set of other substitutions.
  • the substitution L180I appears in variant 3 with P97S, E138A, y194S, A236V, V267I and in variant 18 with N95C, S107D, V167I, G293A, I310K.
  • a variation of the above method is to require (i) that each substitution identified be tried an approximately equal number of times in the designed antibody variant set, and (ii) that as many different combinations of two substitutions (e.g. substitution pairs) as possible be tested.
  • substitution pairs For example, to test forty substitutions in an antibody it may be desirable to incorporate a maximum of five changes per variant. For forty substitutions there are (40 ⁇ 39/2) 780 possible pairs of substitutions. In one variant with five substitutions there are ten pairs of substitutions. So in forty variants there will be a maximum of 400 substitution pairs. The aim is then to maximize the number of different substitution pairs that are tested and to try to represent each substitution five times. The substitution pairs can be scored with the initial selection algorithm, and the top scoring 400 substitution pairs tested.
  • the solution to such a problem of finding variants with the constraints mentioned here is known as a coverage problem.
  • the coverage problem is NP-hard. Therefore greedy and other forms of approximate solutions are used to solve the NP-hard problems in the present invention.
  • the algorithms described in Khan et al., 2001, Lecture Notes in Computer Science 2076: 225 are used.
  • the desired set of sequences can be evolved using monte carlo algorithms and genetic algorithms to maximize the number of pairs in the variant set. Genetic algorithms are described in Section 7.5.1 of Duda et al., 2001 , Pattern Classification , Second Edition, John Wiley & Sons, Inc., New York, which is hereby incorporated by reference in its entirety. Further, similar algorithms can be used to expand the coverage problem to maximize the number of triplets, quadruplets and so on.
  • n the number of identified substitutions
  • n the number of variants to be synthesized
  • k the number of substitutions per variant.
  • ⁇ i Choose two random variants; ii. Choose two random positions; iii. Count the number of distinct substitution pairs seen among variants; iv. Swap the substitutions (if any) at the two positions between the two chosen variants; v. Check if the number of substitutions per variant is k; vi. Check if number of times a given substitution occurs among all variants equals n ⁇ k/m; vii. Count the number of distinct substitution pairs seen among variants; viii. If the count from vii) is greater than count from iii) and v) and vi) are true, accept the changes to the variants from step iv), else, dismiss the changes and retain original values.
  • ⁇ i Choose two random variants; ii. Choose two random positions; iii. Count the number of distinct substitution pairs seen among variants; iv. Swap the substitutions (if any) at the two positions between the two chosen variants; v. Check if the number of substitutions per variant is k;
  • the set of substitutions can be divided into two or more groups and be used to design variants where each variant contains substitutions from a particular group, for example by dividing the antibody into functional domains such as different complementarity determining regions (CDRs) or framework regions.
  • the substitutions in such a variant can be subject the coverage algorithms with constraints described above.
  • Each group can also be combined with other groups of substitutions to design initial variants and coverage algorithm can be applied to combination of substitution groups.
  • Groups of substitutions can be arrived at using knowledge of antibody domain and/or functional and structural properties of amino acid residues in the antibody.
  • substitutions For example we can identify all substitutions based on Section 5.1 and select top scoring ones, and classify them into groups of substitutions based on which domain of the antibody they are present in such as different complementarity determining regions (CDRs) or framework regions. Alternatively, we can also classify substitutions based on the their special location in the protein structure (e.g surface position versus interior positions) based on experimentally determined structure or using prediction algorithms. Alternatively, substitutions can be classified based on their proximity to the binding sites (e.g residues ⁇ 5 ⁇ from the binding site belong to one class and residues >5 ⁇ from the binding site to another).
  • CDRs complementarity determining regions
  • substitutions can be classified based on their proximity to the binding sites (e.g residues ⁇ 5 ⁇ from the binding site belong to one class and residues >5 ⁇ from the binding site to another).
  • Constraints to number of substitutions to be designed in a variant from each substitution group can also be added (e.g., no more than two variants from each substitution group). For example, two substitutions can be chosen from the group close to the binding site and three from the group on the surface of the antibody. Such methods differ from typical experimental design or design of experiment (DOE) methods in the fact that no more than five changes in a variant are allowed and the occurrence of the selected pairs is maximized by scoring. Other DOE methods for distributing 40 substitutions would require as many as between 18 and 22 changes in an antibody, which would have a high likelihood of being detrimental to antibody function.
  • DOE design of experiment
  • an antibody variant set can be created stochastically by library synthesis methods such as parallel site-directed mutagenesis, DNA shuffling or other methods for incorporating defined substitutions into an antibody such as those described in Section 5.8.
  • the variant set contains substitutions distributed at random, so precisely defined variants are not synthesized. Instead, the introduction of substitutions is controlled so that the average number of substitutions incorporated into each variant is between 1 and 10, more preferably the average number of substitutions incorporated into each variant is between 1 and 5.
  • Variants can then be selected at random and the distribution of substitutions can be determined by determining the sequence of the antibody.
  • less than 1000 variants created by library synthesis methods are synthesized and sequenced, preferably less than 500 variants created by library synthesis methods are synthesized and sequenced, more preferably less than 250 variants created by library synthesis methods are synthesized and sequenced, even more preferably less than 100 variants created by library synthesis methods are synthesized and sequenced.
  • the creation of a library can be simulated using computational modeling of shuffling and other methods. See, for example, Moore, 2001, Proc Natl Acad Sci USA 13, 3226-3231; Moore and Maranas, 2000, J Theor Biol. 205, pp. 483-503.
  • the variants are synthesized using methods known in the art. Representative, but nonlimiting synthetic methods are described in Section 5.8, below. Then the antibodies are tested for relevant biological properties. Such relevant biological properties include, but are not limited to antibody solubility and activity. Nonlimiting examples of how such antibody activity can be tested are described in Section 5.9 below. Together the synthesis and testing of the antibody variants represent step 04 in FIG. 2 .
  • step 04 it is desirable to use the sequence and activity information from the designed antibody variant set to assess the contributions of substitutions to the one or more antibody activity or function.
  • This process is represented as step 05 in FIG. 2 .
  • Assessment of the contributions of substitutions to one or more antibody function can be performed by deriving a sequence-activity relationship. Such a relationship can be expressed very generally, for example as shown in Equation 3
  • Y is a quantitative measure of a property of the antibody (e.g., activity),
  • x i is a descriptor of a substitution, a combination of substitutions, or a component of one or more substitutions in the sequence of the antibody, and
  • f( ) is a mathematical function that can take several forms.
  • a model of the sequence-activity relationship can be described as a functional form whose parameters have been trained for the input data (Y and x i ).
  • Protein sequences can be mathematically represented in terms of many variables (descriptors, predictors), each variable representing the type of amino acid at a specific location (linear form in terms of the position of the amino acid).
  • the sequence AGWRY can be represented by five variables, where variable one assumes a value of “A” corresponding to position 1, variable two is “G” corresponding to position two and so on.
  • Each variable can assume 1 of 20 possibilities.
  • a variable can describe position one and two and assume a value of “AG” (thereby creating a variable that in non-linear in terms of position of the amino acid).
  • each position can assume a value corresponding to a physico-chemical property of the amino acid instead of amino acid identity.
  • the position can be described in terms of the mass of the amino acid at that location.
  • a variable for position one can assume the value 71.09 and position two 57.052 and so on.
  • each position can be described by one or several principal components derived to represent many physico-chemical properties of the amino acid present in that position.
  • each variable can be a combination of variables representing properties of amino acids.
  • each variable can be represented in a binary form corresponding to presence or absence of a particular amino acid. For example, consider two variants AGWRY and AKWRY, Position two can be “1 if G is present at that position and “0 if it is absent and the descriptor for that position can have the value “0 or “1.” Alternatively, each variable can be represented in a binary form corresponding to presence or absence of a defined group of amino acids.
  • the functional form f( ) correlates descriptors of an antibody sequence (x i ) to its activity.
  • the function f can be a linear combination of x i :
  • w i is a weight (or coefficients of x i ).
  • a set of descriptors (x i ) that can describe all of the substitutions within the antibody variant set is identified.
  • Values of Y for each member of the antibody variant set are measured.
  • Values for each weight (w i ) are then calculated such that the differences between values predicted for each value of Y by Equation 3 and those observed experimentally are minimized for the antibody variants set, or for a selected subset of such antibody variants.
  • the minimization step above can also use weights for different activity predictions and, in general, can use a loss function.
  • this loss function can be squared error loss, where weights that minimize the sum of squares of the differences between predicted and measured values for the dataset are computed.
  • statistical regression methods are used to identify relationships between dependent (x i ) and independent variables (Y).
  • Such techniques include, but are not limited to, linear regression, non-linear regression, logistic regression, multivariate data analysis, and partial least squares regression. See, for example, Hastie, The Elements of Statistical Learning, 2001, Springer, N.Y.; Smith, Statistical Reasoning, 1985, Allyn and Bacon, Boston.
  • regression techniques like the PLS (Partial Least Square) can be used to solve for the weights (w i ) in the equation X.
  • Partial Least Squares PLS is a tool for modeling linear relationships between descriptors.
  • the method is used to compress the data matrix composed of descriptors (variables) of variant sequences being modeled into a set of latent variable called factors.
  • the number of latent variable is much smaller than the number of variables (descriptors) in the input sequence data. For example, if the number of input variable is 100, the number of latent variables can be less than 10.
  • the factors are determined using the nonlinear iterative partial least squares algorithm.
  • the orthogonal factor scores are used to fit a set of activities to the dependent variables. Even when the predictors are highly collinear or linearly dependent, the method finds a good model.
  • Alternative PLS algorithms like the SIMPLS can also be used for regression. In such methods, the contribution to the activities from every variable can be deconvoluted to study the effect of sequence on the function of the antibody.
  • modeling techniques are used to derive sequence-activity relationships.
  • Such modeling techniques include linear and non-linear approaches. Linear and non-linear approaches are differentiated from each other based on the algebraic relationships used between variables and responses in such approaches.
  • the input data e.g., variables that serve as descriptors of the antibody sequence
  • the input data can be linearly related to the variables provided or non-linear combinations of the variables. It is therefore possible to perform different combinations of models and data-types: linear input variables can be incorporated into a linear model, non-linear input variables can be incorporated into a linear model and non-linear variables can be incorporated into a non-linear models.
  • f( ) Eqn. 3
  • Function fO can assume non-linear form.
  • An example of non-linear functional form is:
  • Non-linear functions can also be derived using modeling techniques such as machine learning methods.
  • sequence (x i )-activity (Y) data to predict the activities of any sequence given the descriptors for a sequence can be determined using neural networks, bayesian models, generalized additive models, support vector machines, classification using regression trees.
  • the data describing variants of the initial antibody can be represented in many forms.
  • all or a portion of the data is represented in a binary format. For example, representing the presence or absence of a specified residue at a particular position by a “1 or a “0 constitutes a linear binary variable. In another example, representing the presence of a specified residue at one position AND a second specified residue at a second position by a “1 constitutes a non-linear binary variable.
  • all or a portion of the data is represented as Boolean operators.
  • all or a portion of the data is represented as principal component descriptors derived from a set of properties. See, for example, Sandberg et al., 1998, J Med. Chem.
  • Antibody input sequence data can also use descriptors based on comparison with a sequence profile (e.g., a hidden Markov model, or principal component analysis of a set of sequences). For example in FIG. 9 , PC 1 and PC 2 values of the sequences can be used as descriptors for the sequences in that Fig. In addition, any number of principle components can be used as descriptors. See, for example, Casari et al., 1995, Nat Struct Biol. 2:171-8; and Gogos et al., 2000, Proteins 40:98-105.
  • step 05 the antibody sequence data in the designed set and the results of the assays performed on the designed set are converted to a form that can be used in pattern classification and/or statistical techniques in order to identify relationships between the results of the assays and the substitutions present in the designed set.
  • conversion involves a step in which independent variables and dependent variables are enumerated.
  • the independent variables are the various substitutions (mutations) that are present in the designed set.
  • the dependent variables are the results of assays, such as those described in Section 5.9.
  • Each substitution can be considered independently.
  • the presence or absence of a substitution or residue at a specific position can be used to describe one or more of the independent variables.
  • the presence or absence of two or more substitutions or residues at two or more specific positions can be used to describe one or more of the independent variables.
  • One or more physico-chemical descriptors of a substitution or residue at a specific position can be used to describe one or more of the independent variables.
  • One or more physico-chemical descriptors of two or more substitutions or residues at two or more specific positions can be used to describe one or more of the independent variables. Then, pattern classification and/or statistical techniques are used to identify relationships between particular substitutions, or combinations of substitutions, and the assay data.
  • supervised learning techniques are used to identify relationships between mutations in the designed set and antibody properties identified in assays results such as assays performed in Section 5.9.
  • Such supervised learning techniques include, but are not limited to, Bayesian modeling, nonparametric techniques (e.g., Parzen windows, k n -Nearest-Neighbor algorithms, and fuzzy classification), neural networks (e.g., hopfield network, multilayer neural networks and support vector machines), and machine learning algorithms (e.g., algorithm-independent machine learning). See, for example, Duda et al., Pattern Classification, 2 nd edition, 2001, John Wiley & Sons, Inc.
  • the sequence (x i )-activity (Y) data can be sed to predict the activities of any sequence given the descriptors for a sequence using a neural network.
  • the input for the network is the descriptors and the output is the predicted value of Y.
  • the weights and the activation function can be trained using supervised decision based learning rules. The learning is performed on a subset of variants called the training set and performance of the network is evaluated on a test set.
  • unsupervised learning techniques are used to identify relationships between mutations in the designed set and antibody properties identified in assays results such as assays performed in Section 5.9.
  • Such unsupervised learning techniques include, but are not limited to stochastic searches (e.g., simulated annealing, Boltzmann learning, evolutionary methods, principal component analysis, and clustering methods). See, for example, Duda et al., Pattern Classification, 2 nd edition, 2001, John Wiley & Sons, Inc. New York.
  • the weights in equation 5 can be adjusted by using monte carlo and genetic algorithms.
  • the optimization of weights for non-linear functions can be complicated and no simple analytical method can provide a good solution in closed form. Genetic algorithms have been successfully used in search spaces of such magnitude. Genetic algorithms and genetic programming techniques can also be used to optimize the function form to best fit the data. For instance, many recombinations of functional forms applied on descriptors of the sequence variants can be applied.
  • boosting techniques are used to construct and/or improve models developed using any of the other techniques described herein.
  • a model of the sequence-activity relationship can be described as a functional form whose parameters have been trained for the input data (Y and x i ).
  • Many algorithms/techniques to build models have been described. Algorithms applied on a specific dataset can be weak in that the predictions can be less accurate or “weak” (yielding poor models). Models can be improved using boosting techniques. See, for example, Hastie et al., The Elements of statistical Learning, 2001, Springer, N.Y.
  • boosting is to combine the outputs of many “weak” predictors into a powerful “committee.”
  • boosting is applied using the AdaBoost algorithm.
  • the prediction algorithm is sequentially applied to repeatedly modified versions of the data thereby producing a sequence of models. The predictions from all of these models are combined through a weighted majority vote to produce the final prediction.
  • the data modification at each step consists of applying weights (W b i ) to each of the i training observations. Initially weights are set to 1/N, where N is the number of training observation (sequence-activity data). The weights are modified individually in each successive iteration.
  • Training observations that were predicted poorly by a particular model have their weights increased and training observations that were predicted more accurately have their weights decreased. This forces each successive model to concentrate on those training observations that are issued by the previous model.
  • the step of combining the models to produce a “committee” assigns a weight to each model based on the overall prediction error of that model.
  • modeling techniques and algorithms described herein can be adapted to derive relationships between one or more desired properties or functions of an antibody and therefore to make multiple predictions from the same model.
  • Modeling techniques that have been adapted to derive sequence-activity relationships for antibodies are within the scope of the present invention. Some of these methods derive linear relationships (for example partial least squares projection to latent structures) and others derive non-linear relationships (for example neural networks). Algorithms that are specialized for mining associations in the data are also useful for designing sequences to be used in the next iteration of sequence space exploration.
  • These modeling techniques can robustly deal with experimental noise in the activity measured for each variant. Often experiments are performed in replicates and for each variant there will be multiple measurement of the same activity.
  • multiple measurements can be averaged and treated as a single number for every variant while modeling the sequence-activity relationship.
  • the average can be a simple mean or another form of an average such as a geometric or a harmonic mean.
  • outliers can be eliminated.
  • the error estimation for a model derived using any algorithm can incorporate the multiple measurements through calculating the standard deviation of the measurement and comparing the predicted activity from the model with the average and estimate the confidence interval within which the prediction lies. Weights for observations to be used in models can also be derived from the accuracy of measurement, for example, through estimating standard deviation and confidence intervals. This procedure can put less emphasis on variants whose measurements are not accurate. Alternatively, theses replicate value can be treated independently.
  • sequence variant i represented by descriptor values ⁇ x j ⁇ i1 has been measured in triplicates (Y i1 , Y i2 , Y i3 )
  • a representative modeling routine in accordance with one embodiment of the invention comprises the following steps.
  • Step 302 Relevant descriptors of the monomeric variables are identified. These descriptors can convey physico-chemical properties relevant to the interaction between biomolecules or classify the monomers (residues) as discreet entities represented in binary form as described earlier. The former is preferred for residue positions in the antibody sequence where the number of different amino acid substitutions is four or more or where the variables can assume one of four possible values for those positions and the physico-chemical properties values are well distributed (e.g.) different from each other. The latter is preferred for positions that have four or less possible values for the relevant variable, and/or the values are clustered (e.g.) are not very different from each other. To create non-linear variables, new variables are formed that are combination of monomeric variables.
  • variable for position 2 assumes a value of “1 if G is present at that position and “0 if it is absent.
  • a non-linear variable can be created in addition to the linear variables describing each position.
  • a new non-linear variable representing position “2 and “3 can assume four values in numeric form.
  • the variable can assume a value of 11 for “GW”, 10 for “GY”, 01 for “KW” and 00 for “KY”.
  • four variables can describe position 2 and 3, where variable one assumes a value of “1 if the sequence at position 2 and 3 is “GW” and “0 otherwise and the second variable takes the values of “1 or “0 if the sequence is “GY” or otherwise and so on.
  • a weight of “1 can be assigned to variables in the heavy chain of the antibody and “0 for variables in light chain of the antibody when modeling activity Y 1 and a weight of “0 can be assigned to variables in heavy chain and “1 for light chain when modeling activity Y 2 .
  • This weighting can also incorporate constraints such as immunogenicity and other functional considerations that may or may not be measured in experiments, but which can be fully or partially predicted using computational techniques.
  • a negative weight can be assigned to appearance of a T-cell epitope in a variant, or removal of glycosylation sites.
  • Step 304 the parameters for the functional form of the sequence-activity relationship are optimized to obtain a model by minimizing the difference between the predicted values and real values of the activity of the antibody.
  • Such optimization adjusts the individual weights for each descriptors identified in preceding steps using a refinement algorithm such as least squares regression techniques.
  • Other methods that use alternative loss functions for minimization can be used to analyze any particular dataset. For example, in some antibody sequence-activity data sets, the activities may not be distributed evenly throughout the measured range. This will skew the model towards data points in the activity space that are clustered. This can be disadvantageous because datasets often contain more data for antibody variants with low levels of activity, so the model or map will be biased towards accuracy for these antibodies that are of lower interest.
  • This skewed distribution can be compensated for by modeling using a probability factor or a cost function based on expert knowledge.
  • This function can be modeled for the activity value or can be used to assign weights to data points based on their activity. As an example, for a set of activities in the range of 0 to 10, transforming the data with a sigmoidal function centered at five will give more weight to sequences with activity above five. Such a function can optionally also be altered with subsequent iterations, thereby focusing the modeling on the part of the dataset with the most desired functional characteristics.
  • This approach can also be coupled with exploring techniques like a Tabu search, where undesired space is explored with lower probabilities.
  • algorithms that optimize the sequence-activity model for the dataset by randomly starting with a solution (e.g., randomly assigning weights w i ) and using methods like hill-descent and/or monte-carlo and/or genetic algorithm approaches to identify optimal solutions.
  • robustness of the models used is a significant criterion.
  • obtaining several sub-optimal solutions from various initial conditions and looking at all the models for common features can be a desirable methodology for ensuring the robustness of the solution.
  • Another way to obtain robust solutions is to create bootstrap data sets based on the input data, than estimate a p-value or confidence on the various coefficients of the model.
  • boosting techniques like AdaBoost can be used to obtain a “committee” based solution.
  • Step 306 Many mathematical modeling techniques for deriving a sequence-activity correlation are evaluated.
  • Preferred mathematical modeling techniques used to identify and capture the sequence-activity correlation handle (i) very large numbers of variables (e.g 20 or more) and correlations between variables, (ii) linear and non-linear interactions between variables, and (iii) are able to extract the variables responsible for a given functional perturbation for subsequent testing of the mathematical model (e.g., models should be easily de-convoluted to assign the effect of variables describing the amino acids substitution with activities).
  • Step 308 the coefficients (parameters) of the model(s) are deconvoluted to see which amino acid substitutions (variables/descriptors of the variants) influence the activity of the antibody. It can be important to identify which descriptor of the antibody are important for the activity of interest.
  • Some of the techniques, such as partial least squares regression (SIMPLS) that uses projection to latent structures (compression of data matrix into orthogonal factors) may be good at directly addressing this point because contributions of variables to any particular latent factors can be directly calculated. See, for example, Bucht et al., 1999, Biochim Biophys Acta. 1431:471-82; and Norinder et al., 1997, J Pept Res 49:155-62.
  • SIMPLS partial least squares regression
  • Modeling techniques/methods that directly correlate the amino acid variations to the activity are preferred because we can derive the sequence-activity map (relationship) to construct new variants not in dataset that have preferentially higher activities. These methods can be adapted to provide a direct answer and output in desired forms.
  • Step 310 the models developed using various algorithms and methods in the previous step can be evaluated by cross validation methods. For example, by randomly leaving data out to build a model and making predictions of data not incorporated into the model is a standard technique for cross validation. In some instances of antibody engineering, data may be generated over a period of months. The data can be added incrementally to the modeling procedure as and when such data becomes available. This can allow for validation of the model with partial or additional datasets, as well as predictions for the properties of antibody sequences for which activities are still not available. This information may then be used to validate the model.
  • FIGS. 4 and 5 An example of internal model validation methods is shown in FIGS. 4 and 5 .
  • a confidence score for each regression coefficient or weight vector can be generated for any antibody sequence-activity model.
  • average values for weight functions can be obtained by omitting a part of the available data. Either individual sequences and their associated activities or individual substitution positions can be left out. A sequence-activity relationship can then be constructed from this partial data. This process can be repeated many times, each time the data to leave out is selected randomly. Finally an average and range of values for each weight function is calculated. The weight functions can then also be ranked in order of their importance to activity.
  • Step 312 new antibody sequences that are predicted to possess one or more desired property are derived. Alternatively it can be desirable to rank order the input variables for detailed sequence-activity correlation measures.
  • the model can be used to propose sequences that have high probabilities of being improved. Such sequences can then be synthesized and tested. In one embodiment, this can be achieved if the effects of various sequence features of the antibodies on their functions are known based on the modeling. Alternatively, for methods like neural networks, 10 3 or 10 6 or 10 9 or 1012 or 10 15 or 10 18 or as many as 10 80 sequences can be evaluated in silico. Then those predicted by the model to possess one or more desired properties are selected.
  • Step 314 The statistical quality of the model fit to the input data is evaluated in step 314 .
  • Validation of sequence-activity correlation can be internal, using cross-validation of the data, or preferably external, by forecasting the functional perturbation of a set of new sequences derived from the model. Sequences with predicted values of their functional perturbations are then physically made and tested in the same experimental system used to quantify the training set. If the sequence-activity relationship of the dataset is satisfactory quantified using internal and external validation, the model can be applied to a) predict the functional value of other related sequences not present in the training set, and b) design new sequences within the described space that are likely to have a function value that is outside or within the range of function given by the training set.
  • the initial set of data can be small, so models built from it can be inaccurate. Initial models may not contain terms to account for amino acid interactions. Others have found that amino acid changes within an antibody are approximately additive and few interaction terms are required to describe the effects of mutations on protein function. See, for example, Aita et al. (2000) Biopolymers 54: 64-79.; Aita et al. (2001) Protein Eng 14: 633-8.; Choulier et al. (2002) Protein Eng 15: 373-82.; and Prusis et al. (2002) Protein Eng 15: 305-11. However such interactions can be important and can result in a variant that incorporates all beneficial changes having low activity (Aita et al. (2002) Antibodies 64: 95-105.). Improving the modeled relationship further depends upon obtaining better values for weights whose confidence scores are low. To obtain this data, additional variants designed as described in Section 5.4 below will provide additional data useful in establishing more precise sequence-activity relationships.
  • the output from each method for modeling a sequence-activity relationship can be one or more of: (i) a regression coefficient, weight or other value describing the relative or absolute contribution of each substitution or combination of substitutions to one or more activity of the antibody, (ii) a standard deviation, variance or other measure of the confidence with which the value describing the contribution of the substitution or combination of substitutions to one or more activity of the antibody can be assigned, (iii) a rank order of preferred substitutions, (iv) the additive & non-additive components of each substitution or combination of substitutions, (v) a mathematical model that can be used for analysis and prediction of the functions of in silico generated sequences, (vi) a modification of one or more inputs or weights used by an expert system 100 to select substitutions or (vii) a modification of the methods used by expert system 100 to design an antibody variant set.
  • each different method for deriving relationships between antibody sequences and activities can differ in the precise values of their outputs. In some embodiments of the invention it is therefore desirable to combine the outputs from two or more such methods for subsequent uses. This corresponds to step 06 in FIG. 2 . There are a variety of ways in which such outputs can be combined. In some embodiments, each output can be independently applied to the subsequent design of antibody variants ( FIG. 2 , step 07 ) or the modification of parameters or weights used by expert system 100 for the selection of substitutions ( FIG. 2 step 02 ) or the design of antibody variant sets ( FIG. 2 step 03 ).
  • average values can be calculated for the regression coefficient, weight or other value describing the relative or absolute contribution of each substitution or combination of substitutions to one or more activity of the antibody (e.g., as defined in Equation 4 below).
  • the standard deviation, variance or other measure of the confidence with which the value describing the contribution of the substitution or combination of substitutions to one or more activity of the antibody can be assigned (e.g., as defined in Equation 4 below).
  • the rank order of preferred substitutions is used to combine the methods.
  • the additive (linear variables) and non-additive components (non-linear variables) of each substitution or combination of substitutions is combined:
  • V ix f ( M 1 ( i x ), M 2 ( i x ), ⁇ M j ( i x )) (Eq. 6)
  • V ix is a combined measure of one of the descriptors measuring the performance of an antibody in which monomer x is substituted at position i;
  • M j (i x ) is a measure of one of descriptors measuring the performance of an antibody in which monomer x is substituted at position i, determined by sequence-activity correlating method j(M j (i x ) is the contribution of i x as determined by Model j)
  • f( ) is some mathematical function.
  • f( ) can be a linear combination of contribution of i x from many models.
  • independent values can be obtained for the functional values of in silico generated sequences derived from two or more mathematical models by using the model generated in the prior steps to predict/calculate the value of the new sequence represented in terms of the same variables that are used to build the model.
  • average values (or some other mathematical function of two or more values derived by two or more sequence-activity models) can be obtained for the functional values of in silico generated sequences derived from two or more mathematical models.
  • sequence-activity relationships can be chosen or modified such that they better predict the performance of individual substitutions within a combination of other substitutions in an antibody, as described in more detail in Subsection 5.4.4.
  • step 07 of FIG. 2 This corresponds to step 07 of FIG. 2 .
  • this step is similar to the processes corresponding to steps 02 and 03 in FIG. 2 . It involves defining a sequence space in terms of an antibody sequence and a set of substitutions, then designing a set of antibody variants that incorporate those substitutions in different combinations.
  • sequence space can be defined in terms of the original target antibody sequence and substitutions that have already been tested.
  • this method for defining the sequence space is used if the desired degree of further increase in one or more activity of the antibody is less than 10-fold, preferably less than 5-fold, more preferably less than 2-fold.
  • sequence space can be defined in terms of the original target antibody sequence and a combination of substitutions that have already been tested and those that have not yet been tested.
  • this method for defining the sequence space is used if the desired degree of further increase in one or more activity of the antibody is greater than 2-fold, preferably greater than 5-fold, and more preferably greater than 10-fold.
  • sequence space can be defined purely in terms of the original target antibody sequence and substitutions that have not yet been tested. This method for defining the sequence space is generally most appropriate for the initial variant set as represented in FIG. 2 step 02 .
  • substitutions can also mean a pair or larger group of substitutions (for example, when the descriptors of antibodies are represented in non-linear form as described in section 5.3), since sequence-activity relationships can produce regression coefficients, weights or other measurements of contribution to function and confidences for these measurements that apply not to individual substitutions but to specific combinations of these substitutions.
  • a substitution can be selected if it has a positive regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody.
  • a substitution can be selected if it has a positive regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it is at least one standard deviation, preferably two standard deviations or more preferably three standard deviations above neutrality.
  • a substitution can be selected if it has a positive regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it has also been tested at least once, preferably at least twice, more preferably at least 3 times, more preferably at least 4 times, even more preferably at least 5 times.
  • a substitution can be selected if it has a positive regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it is at least one standard deviation, preferably two standard deviations or more preferably three standard deviations above neutrality, and it has also been tested at least once, preferably at least twice, more preferably at least 3 times, more preferably at least 4 times, even more preferably at least 5 times.
  • a substitution can be selected from a rank ordered list of substitutions. For example the most favorable substitution may be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 or 20 most favorable substitutions can be selected.
  • a substitution can be selected from a rank ordered list of substitutions.
  • the most favorable substitution can be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 or 20 most favorable substitutions can be selected, and it is at least one standard deviation, preferably two standard deviations or more preferably three standard deviations above neutrality.
  • a substitution can be selected from a rank ordered list of substitutions.
  • the most favorable substitution can be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 or 20 most favorable substitutions can be selected, and it has also been tested at least once, preferably at least twice, more preferably at least 3 times, more preferably at least 4 times, even more preferably at least 5 times.
  • a substitution can be selected from a rank ordered list of substitutions.
  • the most favorable substitution may be selected, or the 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15, 16, 17, 18, 19 or 20 most favorable substitutions can be selected, and it is at least one standard deviation, preferably two standard deviations or more preferably three standard deviations above neutrality, and it has also been tested at least once, preferably at least twice, more preferably at least 3 times, more preferably at least 4 times, even more preferably at least 5 times.
  • a substitution can be selected if it has a negative regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it is less than three standard deviations, preferably less than two standard deviations or more preferably less than one standard deviation below neutrality.
  • a substitution can be selected if it has a negative regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it has also been tested no more than 5 times, preferably no more than 4 times, more preferably no more than 3 times, more preferably no more than twice, even more preferably no more than once.
  • a substitution can be selected if it has a negative regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it is less than three standard deviations, preferably less than two standard deviations or more preferably less than one standard deviation below neutrality, and it has also been tested no more than 5 times, preferably no more than 4 times, more preferably no more than 3 times, more preferably no more than twice, even more preferably no more than once.
  • substitution can be eliminated if it has a negative regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody.
  • a substitution can be eliminated if it has a negative regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it is at least one standard deviation, preferably two standard deviations or more preferably three standard deviations above neutrality.
  • a substitution can be eliminated if it has a negative regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it has also been tested at least once, preferably at least twice, more preferably at least 3 times, more preferably at least 4 times, even more preferably at least 5 times.
  • a substitution can be eliminated if it has a negative regression coefficient, weight or other value describing its relative or absolute contribution to one or more activity of the antibody, and it is at least one standard deviation, preferably two standard deviations or more preferably three standard deviations above neutrality, and it has also been tested at least once, preferably at least twice, more preferably at least 3 times, more preferably at least 4 times, even more preferably at least 5 times.
  • Antibody variants that combine or eliminate previously tested substitutions can serve at least two purposes. First, they can be used to obtain antibody variants that are improved for one or more property, activity or function of interest. Generally, though not exclusively, substitutions selected according to criteria (i)-(viii) in subsection 5.4.2 are most likely to be appropriate for this purpose. Second, they can be used to obtain additional information relating the sequence to the activity of an antibody, thereby improving the accuracy with which predictions can be made concerning the effect of substitutions upon one or more property, activity or function of an antibody. Generally, though not exclusively, substitutions selected according to criteria (i)-(xi) in subsection 5.4.2 are most likely to be appropriate for this purpose.
  • Method 1 An antibody that has previously been tested for the one or more property, activity or function of interest is selected.
  • the selected antibody has one of the 100 highest experimentally measured scores for the property, activity or function of interest, more preferably one of the 50 highest experimentally measured scores, even more preferably one of the 25 highest experimentally measured scores, even more preferably one of the 10 highest experimentally measured scores.
  • substitutions in the selected antibody are combined with one or more substitutions selected by one or more of the methods described in subsection 5.4.2.
  • less than 10 selected substitutions are used, more preferably less than 5 selected substitutions are used, even more preferably less than 3 selected substitutions are used.
  • Method 2 An antibody that has previously been tested for the one or more property, activity or function of interest is selected.
  • the selected antibody has one of the 100 highest experimentally measured scores for the property, activity or function of interest, more preferably one of the 50 highest experimentally measured scores, even more preferably one of the 25 highest experimentally measured scores, even more preferably one of the 10 highest experimentally measured scores.
  • substitutions in the selected antibody are combined with one or more substitutions selected by one or more of the methods described in subsection 5.4.2.
  • less than 10 selected substitutions are used, more preferably less than 5 selected substitutions are used, even more preferably less than 3 selected substitutions are used.
  • these substitutions are combined with one or more substitutions selected by one or more method described in Section 5.1 (i.e., by the methods used in step 03 of FIG. 2 ).
  • less than 10 of these last selected substitutions are used, more preferably less than 5 of these last selected substitutions are used, even more preferably less than 3 of these last selected substitutions are used.
  • Method 3 Two or more substitutions identified by one or more of the methods described in subsection 5.4.2 are selected. In preferred embodiments less than 100 selected substitutions, more preferably less than 50, and even more preferably less than 25 are used. One or more antibody variants containing these substitutions are designed using the methods described in Section 5.2.
  • Method 4 One or more substitutions selected by one or more of the methods described in subsection 5.4.2 are selected. In preferred embodiments less than 100 selected substitutions, more preferably less than 50, and even more preferably less than 25 are used. One or more substitutions are selected using one or more of the methods described in Section 5.1. In preferred embodiments, less than 100, and more preferably less than 50 of these selected substitutions are used. Then, one or more antibody variants are designed using the methods described in Section 5.2.
  • Method 5 One or more substitutions selected by one or more of the methods described in subsection 5.4.2 that contribute most positively to the property (e.g., function, activity of interest) are selected. In preferred embodiments, between 1 and 20 most positive substitutions are selected. One or more antibody variant that has already been tested for the property is selected. In preferred embodiments, the between 1 and 20 most active antibodies are selected. One or more of the selected substitutions is added to each of the one or more selected antibodies. In preferred embodiments, the number of substitution positions to be added to each antibody variant sequence is between 1 and 10, more preferably between 1 and 6, and even more preferably between 1 and 3.
  • Substitutions whose regression coefficients, weights or other values describing the relative or absolute contribution to one or more activity of the antibody are positive are selected. Those substitutions whose regression coefficients, weights or other values describing the relative or absolute contribution to one or more activity of the antibody have confidences within a threshold distance from the randomized average weight for that substitution are eliminated. In preferred embodiments, this threshold distance is within 1 standard deviation, more preferably within 2 standard deviations.
  • the substitutions with positive weights and high confidences are combined into a single variant. Alternatively, the selected substitutions are used to design a set of antibody variants as described in Section 5.2.
  • substitutions are ranked in the order in which confidences can be assigned to regression coefficients, weights or other values describing the relative or absolute contribution to one or more activity of the antibody.
  • the substitutions with lowest confidence scores are selected. From the sequences of antibody variants whose activities have already been measured, those that have high values for the property of interest are selected. In preferred embodiments, between 1 and 20 tested antibody variant sequences with highest activities are selected.
  • One or more of the selected substitutions is added to each selected variant. In preferred embodiments, the number of substitutions to be added to each antibody variant sequence is between 1 and 10, more preferably between 1 and 6, and even more preferably between 1 and 3.
  • Method 8 One or more antibody variants that have already been tested for the property of interest are selected. In preferred embodiments, between 1 and 20 most active antibodies are selected. One or more substitutions for which a contribution to the property has been calculated are selected. For each of the one or more selected antibodies, the following process is performed. One of the selected substitutions is added or removed and the predicted activity of the resultant antibody is calculated using one or more models for sequence-activity relationship as described in the section 5.3.
  • Exemplary models include, but are not limited to (i) regression techniques that provide regression coefficients for the descriptors, (ii) models that generate weights or other value describing the relative or absolute contribution of each substitution or combination of substitutions to one or more activity of the antibody, (iii) models that provide standard deviation, variance or other measures of the confidence with which the value describing the contribution of the substitution or combination of substitutions to one or more activity of the antibody can be assigned, (iv) models that rank order preferred substitutions, (v) models that provide additive and non-additive components of each substitution or combination of substitutions, (vi) analytical mathematical models that can be used for analysis and prediction of the functions of in silico generated sequences (vii) supervised and unsupervised machine learning techniques like neural networks that can predict the activity of new antibody sequences expressed in terms of the descriptors that are used in modeling.
  • the process reverts to the sequence of the antibody before the change. This process continues for a certain number of steps (preferably more than 10 steps, more preferably more than 100 steps, even more preferably more than 1000 steps) or until the predicted activity of the antibody converges to a value. Either the final antibody sequence in the series of iterations of the method, or the antibody sequence in the series with the highest predicted activity is selected. This process can optionally be performed more than once starting from each initial antibody sequence.
  • Method 9 As an optional addition to any of the design methods including methods 1, 2, 5, and 7, one or more substitutions determined to be detrimental to the desired property (e.g., by any of the criteria described in subsection 5.4.2 including criteria (xii)-(xv)) are eliminated.
  • Method 10 As an optional addition to any design method, newly designed variants that can be reached by making a certain number of substitutions to an antibody sequence whose activity has already been measured are discarded and not synthesized. In preferred embodiments newly designed variants that can be reached by making 10 or fewer substitutions to an antibody sequence whose activity has already been measured are not synthesized. More preferably, newly designed variants that can be reached by making 5 or fewer substitutions to an antibody sequence whose activity has already been measured are not synthesized. More preferably, newly designed variants that can be reached by making 3 or fewer substitutions to an antibody sequence whose activity has already been measured are not synthesized.
  • newly designed variants that can be reached by making 2 or fewer substitutions to an antibody sequence whose activity has already been measured are not synthesized.
  • newly designed variants that can be reached by making 1 to an antibody sequence whose activity has already been measured are not synthesized.
  • sequence-activity modeling methods can be quantitatively compared. Such comparisons can be used to modify variable parameters within each method, or to select methods of combining the results of two or more sequence-activity correlating methods as outlined in Subsection 5.3.1.
  • the outputs of methods that determine sequence-activity relationship are outlined in Section 5.3. These outputs can be combined to calculate the predicted activity of an antibody and the confidence with which that activity can be predicted. These predictions can be compared with activity values obtained experimentally for newly designed and synthesized antibody variants, and the method or methods of deriving sequence-activity relationships may be chosen or modified in one or more of the following ways.
  • weights applied to the scores produced by the one or more sequence-activity correlating methods for example as shown in Equation 4 or as described in Subsection 5.3.1 can be modified such that one or more of the following are true.
  • sequence-activity correlating method is chosen such that one or more of the following are true.
  • steps 1 or 2 can be performed using regression techniques, machine learning or other multivariate data analysis tools to calculate or minimize the differences between the values predicted by the sequence-activity relationship, and those observed experimentally.
  • steps 1 or 2 can be performed using values predicted by the sequence-activity relationship, and those observed experimentally for more than one set of antibodies.
  • the process of step 4 can be performed using two or more datasets from antibodies that fall into the same class and subclass.
  • Weights for expert system rules 120 that are modified using two or more datasets from antibodies of the same class and subclass can be stored, for example in knowledge base 108 or case-specific data 110 . These weights or choices for sequence-activity determining methods can then be used by expert system 100 when a subsequent target antibody sequence and activity dataset of that class and subclass is presented.
  • step 4 can be performed using two or more datasets from antibodies that fall into the same class. For example two or more sets of human antibodies, two or more sets of murine antibodies. Weights for expert system 100 rules 120 that are modified using two or more datasets from antibodies of the same class can be stored, for example in knowledge base 108 or case-specific data 110 . These weights for expert system 100 rules 120 can then be used by expert system 100 when a subsequent target antibody sequence and activity dataset of that class and subclass is presented.
  • the endpoint of a process of antibody optimization is reached when one or more antibodies are obtained with one or more properties at the levels defined by a user, these activity levels being appropriate to allow the use of the antibody in performing a specific task. This corresponds to FIG. 2 step 08 .
  • information from sequence-activity relationships can be used to provide information to improve the initial selection of substitutions, for example by modifying the weights applied to the scores produced by the expert system 100 as described in Section 5.1.
  • the weights can be modified according to the following process.
  • the sequence-activity relationship can be used to calculate (i) a regression coefficient, weight or other value describing the relative or absolute contribution of each substitution or combination of substitutions to one or more activity of the antibody, (ii) a standard deviation, variance or other measure of the confidence with which the value describing the contribution of the substitution or combination of substitutions to one or more activity of the antibody can be assigned, and/or (iii) a rank order of preferred substitutions.
  • the results of applying two or more rules 120 of expert system 100 are combined and can be used to obtain (i) a score describing the predicted effect of a substitution upon one or more antibody property, (ii) a probability or confidence describing the predicted effect of a substitution upon one or more antibody property, activity or function, or (iii) a predicted rank order of preferred substitutions.
  • Different values for each of these predictions can result from modifications of the weights applied to the scores produced by expert system 100 as described in Section 5.1, for example as shown in equations (1) or (2).
  • the weights applied to the scores produced by expert system 100 can be modified such that one or more of the following are true.
  • steps 1 to 3 can be performed using regression techniques, machine learning or other multivariate data analysis tools to minimize the differences between the values obtained from the sequence-activity relationship, and those predicted by expert system 100 .
  • steps 1 to 3 can be performed using expert system 100 predictions and sequence-activity relationships for more than one set of antibodies.
  • the process of step 5 can be performed using two or more datasets from antibodies that fall into the same class and subclass.
  • Weights for expert system 100 rules 120 that are modified using two or more datasets from antibodies of the same class and subclass can be stored, for example in knowledge base 108 or case-specific data 110 . These weights for expert system rules 120 can then be used by expert system 100 when a subsequent target antibody of that class and subclass is presented.
  • step 5 can be performed using two or more datasets from antibodies that fall into the same class. For example, two or more sets of human antibodies, two or more sets of murine antibodies. Weights for expert system 100 rules 120 that are modified using two or more datasets from antibodies of the same class can be stored, for example in knowledge base 108 or case-specific data 110 . These weights for rules 120 can then be used by expert system 100 when a subsequent target antibody of that class is presented.
  • predictions made by expert system 100 can be improved so that preferences (e.g. higher weights) are given to selection methods 130 that have performed well in previous iterations.
  • Different algorithms and methods for identifying productive substitutions and for deriving sequence activity relationships may be better suited to different types of antibody, including different animal origins, different antibody fragments, optimization compared with humanization.
  • Such a computational system could be made available directly, via the internet and/or on a subscription basis.
  • antibodies incorporating substitutions identified through construction and characterizing sets of variant antibodies include antibodies incorporating substitutions identified through construction and characterizing sets of variant antibodies.
  • this also includes vectors (including expression vectors) comprising such polynucleotides, host cells comprising such polynucleotides and/or vectors, and libraries of antibodies, and libraries of host cells comprising and/or expressing such libraries of antibodies.
  • the antibodies developed using the methods of the invention can be used alone or in combination with other prophylactic or therapeutic agents for treating, managing, preventing or ameliorating a disorder or one or more symptoms thereof.
  • the present invention provides methods for preventing, managing, treating, or ameliorating a disorder comprising administering to a subject in need thereof one or more antibodies of the invention alone or in combination with one or more therapies (e.g., one or more prophylactic or therapeutic agents) other than an antibody of the invention.
  • the present invention also provides compositions comprising one or more antibodies of the invention and one or more prophylactic or therapeutic agents other than antibodies of the invention and methods of preventing, managing, treating, or ameliorating a disorder or one or more symptoms thereof utilizing said compositions.
  • Therapeutic or prophylactic agents include, but are not limited to, small molecules, synthetic drugs, peptides, polypeptides, proteins, nucleic acids (e.g., DNA and RNA nucleotides including, but not limited to, antisense nucleotide sequences, triple helices, RNAi, and nucleotide sequences encoding biologically active proteins, polypeptides or peptides) antibodies, synthetic or natural inorganic molecules, mimetic agents, and synthetic or natural organic molecules.
  • nucleic acids e.g., DNA and RNA nucleotides including, but not limited to, antisense nucleotide sequences, triple helices, RNAi, and nucleotide sequences encoding biologically active proteins, polypeptides or peptides
  • synthetic or natural inorganic molecules e.g., synthetic drugs, peptides, polypeptides, proteins, nucleic acids (e.g., DNA and RNA nucleotides including, but not limited to
  • Any therapy that is known to be useful, or that has been used or is currently being used for the prevention, management, treatment, or amelioration of a disorder or one or more symptoms thereof can be used in combination with an antibody of the invention in accordance with the invention described herein. See, e.g., Gilman et al., Goodman and Gilman's: The Pharmacological Basis of Therapeutics, 10th ed., McGraw-Hill, New York, 2001; The Merck Manual of Diagnosis and Therapy, Berkow, M. D. et al.
  • agents include, but are not limited to, immunomodulatory agents, anti-inflammatory agents (e.g., adrenocorticoids, corticosteroids (e.g., beclomethasone, budesonide, flunisolide, fluticasone, triamcinolone, methlyprednisolone, prednisolone, prednisone, hydrocortisone), glucocorticoids, steroids, non-steriodal anti-inflammatory drugs (e.g., aspirin, ibuprofen, diclofenac, and COX-2 inhibitors), pain relievers, leukotreine antagonists (e.g., montelukast, methyl xanthines, zafirlukast, and zileuton), beta2-agonists (e.g., albuterol, biterol, fenoterol, isoetharie, metaproterenol, pirbuterol, salbutamol, terbutalin for
  • antibodies of the invention can be used directly against a particular antigen.
  • antibodies of the invention belong to a subclass or isotype that is capable of mediating the lysis of cells to which the antibody binds.
  • the antibodies of the invention belong to a subclass or isotype that, upon complexing with cell surface proteins, activates serum complement and/or mediates antibody dependent cellular cytotoxicity (ADCC) by activating effector cells such as natural killer cells or macrophages.
  • ADCC antibody dependent cellular cytotoxicity
  • antibodies The biological activities of antibodies are known to be determined, to a large extent, by the constant domains or Fc region of the antibody molecule (Uananue and Benacerraf, Textbook of Immunology, 2nd Edition, Williams & Wilkins, p. 218 (1984)). This includes their ability to activate complement and to mediate antibody-dependent cellular cytotoxicity (ADCC) as effected by leukocytes.
  • ADCC antibody-dependent cellular cytotoxicity
  • Antibodies of different classes and subclasses differ in this respect, as do antibodies from the same subclass but different species; according to the present invention, antibodies of those classes having the desired biological activity are prepared.
  • mouse antibodies of the IgG2a and IgG3 subclass and occasionally IgG1 can mediate ADCC
  • antibodies of the IgG3, IgG2a, and IgM subclasses bind and activate serum complement.
  • Complement activation generally requires the binding of at least two IgG molecules in close proximity on the target cell. However, the binding of only one IgM molecule activates serum complement.
  • any particular antibody to mediate lysis of the target cell by complement activation and/or ADCC can be assayed.
  • the cells of interest are grown and labeled in vitro; the antibody is added to the cell culture in combination with either serum complement or immune cells which may be activated by the antigen antibody complexes. Cytolysis of the target cells is detected by the release of label from the lysed cells.
  • antibodies can be screened using the patient's own serum as a source of complement and/or immune cells. The antibody that is capable of activating complement or mediating ADCC in the in vitro test can then be used therapeutically in that particular patient.
  • IgM antibodies may be preferred for certain applications, however IgG molecules by being smaller may be more able than IgM molecules to localize to certain types of infected cells.
  • the antibodies of this invention are useful in passively immunizing patients.
  • the antibodies of the invention can also be used in diagnostic assays either in vivo or in vitro for detection/identification of the expression of an antigen in a subject or a biological sample (e.g., cells or tissues).
  • a biological sample e.g., cells or tissues.
  • Non-limiting examples of using an antibody, a fragment thereof, or a composition comprising an antibody or a fragment thereof in a diagnostic assay are given in U.S. Pat. Nos.
  • Non-limiting examples are an ELISA, sandwich assay, and steric inhibition assays.
  • the antibodies may be conjugated to a label that can be detected by imaging techniques, such as X-ray, computed tomography (CT), ultrasound, or magnetic resonance imaging (MRI).
  • CT computed tomography
  • MRI magnetic resonance imaging
  • the antibodies of the invention can also be used for the affinity purification of the antigen from recombinant cell culture or natural sources.
  • nucleic acids are written left to right in 5′ to 3′ orientation; amino acid sequences are written left to right in amino to carboxy orientation, respectively.
  • the headings provided herein are not limitations on the invention, but exemplify the various aspects of the invention. Accordingly, the terms defined immediately below are more fully defined by reference to the specification as a whole.
  • polynucleotide oligonucleotide
  • nucleic acid and “nucleic acid molecule” and “gene” are used interchangeably herein to refer to a polymeric form of nucleotides of any length, and may comprise ribonucleotides, deoxyribonucleotides, analogs thereof, or mixtures thereof. This term refers only to the primary structure of the molecule. Thus, the term includes triple-, double- and single-stranded deoxyribonucleic acid (“DNA”), as well as triple-, double- and single-stranded ribonucleic acid (“RNA”).
  • DNA triple-, double- and single-stranded deoxyribonucleic acid
  • RNA triple-, double- and single-stranded ribonucleic acid
  • polynucleotide oligonucleotide
  • nucleic acid containing D-ribose
  • polyribonucleotides including tRNA, rRNA, HRNA, siRNA and mRNA, whether spliced or unspliced, any other type of polynucleotide which is an N- or C-glycoside of a purine or pyrimidine base, and other polymers containing nonnucleotidic backbones, for example, polyamide (e.g., peptide nucleic acids (“PNAs”)) and polymorpholino (commercially available from the Anti-Virals, Inc., Corvallis, Oreg., as Neu
  • PNAs peptide nucleic acids
  • these terms include, for example, 3′-deoxy-2′,5′-DNA, oligodeoxyribonucleotide N3′ P5′ phosphoramidates, 2′-O-alkyl-substituted RNA, double- and single-stranded DNA, as well as double- and single-stranded RNA, and hybrids thereof including for example hybrids between DNA and RNA or between PNAs and DNA or RNA, and also include known types of modifications, for example, labels, alkylation, “caps,” substitution of one or more of the nucleotides with an analog, internucleotide modifications such as, for example, those with uncharged linkages (e.g., methyl phosphonates, phosphotriesters, phosphoramidates, carbamates, etc.), with negatively charged linkages (e.g., phosphorothioates, phosphorodithioates, etc.), and with positively charged linkages (e.g., aminoalkylphosphoramidates, amino
  • nucleases nucleases
  • toxins antibodies
  • signal peptides poly-L-lysine, etc.
  • intercalators e.g., acridine, psoralen, etc.
  • chelates of, e.g., metals, radioactive metals, boron, oxidative metals, etc.
  • alkylators those with modified linkages (e.g., alpha anomeric nucleic acids, etc.), as well as unmodified forms of the polynucleotide or oligonucleotide.
  • nucleotides which can perform that function or which can be modified (e.g., reverse transcribed) to perform that function are used.
  • nucleotides are to be used in a scheme which requires that a complementary strand be formed to a given polynucleotide, nucleotides are used which permit such formation.
  • nucleoside and nucleotide will include those moieties which contain not only the known purine and pyrimidine bases, but also other heterocyclic bases which have been modified. Such modifications include methylated purines or pyrimidines, acylated purines or pyrimidines, or other heterocycles. Modified nucleosides or nucleotides can also include modifications on the sugar moiety, e.g., wherein one or more of the hydroxyl groups are replaced with halogen, aliphatic groups, or are functionalized as ethers, amines, or the like.
  • the term “nucleotidic unit” is intended to encompass nucleosides and nucleotides.
  • modifications to nucleotidic units include rearranging, appending, substituting for or otherwise altering functional groups on the purine or pyrimidine base which form hydrogen bonds to a respective complementary pyrimidine or purine.
  • the resultant modified nucleotidic unit optionally may form a base pair with other such modified nucleotidic units but not with A, T, C, G or U. Abasic sites may be incorporated which do not prevent the function of the polynucleotide. Some or all of the residues in the polynucleotide can optionally be modified in one or more ways.
  • Standard A-T and G-C base pairs form under conditions which allow the formation of hydrogen bonds between the N3-H and C4-oxy of thymidine and the N1 and C6-NH2, respectively, of adenosine and between the C2-oxy, N3 and C4-NH2, of cytidine and the C2-NH2, N′—H and C6-oxy, respectively, of guanosine.
  • guanosine (2-amino-6-oxy-9-.beta.-D-ribofuranosyl-purine) may be modified to form isoguanosine (2-oxy-6-amino-9-.beta.-D-ribofuranosyl-purine).
  • isocytidine may be prepared by the method described by Switzer et al. (1993) Biochemistry 32:10489-10496 and references cited therein; 2′-deoxy-5-methyl-isocytidine may be prepared by the method of Tor et al. (1993) J. Am. Chem. Soc. 115:4461-4467 and references cited therein; and isoguanine nucleotides may be prepared using the method described by Switzer et al. (1993), supra, and Mantsch et al. (1993) Biochem. 14:5593-5601, or by the method described in U.S. Pat. No. 5,780,610 to Collins et al.
  • DNA sequence refers to a contiguous nucleic acid sequence.
  • the sequence can be either single stranded or double stranded, DNA or RNA, but double stranded DNA sequences are preferable.
  • the sequence can be an oligonucleotide of 6 to 20 nucleotides in length to a full length genomic sequence of thousands of base pairs.
  • proteins refers to contiguous “amino acids” or amino acid “residues.” Typically, proteins have a function. However, for purposes of this invention, proteins also encompasses polypeptides and smaller contiguous amino acid sequences that do not have a functional activity. “Polypeptide” and “protein” are used interchangeably herein and include a molecular chain of amino acids linked through peptide bonds. The terms do not refer to a specific length of the product. Thus, “peptides,” “oligopeptides,” and “proteins” are included within the definition of polypeptide.
  • polypeptides containing in co- and/or post-translational modifications of the polypeptide made in vivo or in vitro for example, glycosylations, acetylations, phosphorylations, PEGylations and sulphations.
  • protein fragments, analogs including amino acids not encoded by the genetic code, e.g. homocysteine, ornithine, p-acetylphenylalanine, D-amino acids, and creatine
  • natural or artificial mutants or variants or combinations thereof fusion proteins, derivatized residues (e.g. alkylation of amine groups, acetylations or esterifications of carboxyl groups) and the like are included within the meaning of polypeptide.
  • amino acids or “amino acid residues” may be referred to herein by either their commonly known three letter symbols or by the one-letter symbols recommended by the IUPAC-IUB Biochemical Nomenclature Commission. Nucleotides, likewise, may be referred to by their commonly accepted single-letter codes.
  • Sequence variants refers to variants of discrete antibodies (that is antibodies whose sequence can be uniquely defined) including polynucleotide and polypeptide and variants. Sequence variants are sequences that are related to one another or to a common nucleic acid or amino acid “reference sequence” but contain some differences in nucleotide or amino acid sequence from each other. These changes can be transitions, transversions, conservative substitutions, non-conservative substitutions, deletions, insertions or substitutions with non-naturally occurring nucleotides or amino acids (mimetics).
  • the phrase “optimizing a sequence” refers to the process of creating nucleic acid or protein variants so that the desired functionality and or properties of the protein or nucleic acid are improved. One of skill will realize that optimizing an antibody could involve selecting a variant with lower functionality than the parental protein if that is desired.
  • antibody as used herein includes antibodies obtained from both polyclonal and monoclonal preparations, as well as: hybrid (chimeric) antibody molecules (see, for example, Winter et al. (1991) Nature 349:293-299; and U.S. Pat. No. 4,816,567); F(ab′)2 and F(ab) fragments; Fv molecules (noncovalent heterodimers, see, for example, Inbar et al. (1972) Proc Natl Acad Sci USA 69:2659-2662; and Ehrlich et al. (1980) Biochem 19:4091-4096); single-chain Fv molecules (sFv) (see, for example, Huston et al.
  • antibody and “antibodies” further refer to monoclonal antibodies, multispecific antibodies, human antibodies, humanized antibodies, camelised antibodies, chimeric antibodies, single-chain Fvs (scFv), single chain antibodies, single domain antibodies, Fab fragments, F(ab) fragments, disulfide-linked Fvs (sdFv), anti-idiotypic (anti-Id) antibodies, and epitope-binding fragments of any of the above.
  • antibodies include immunoglobulin molecules and immunologically active fragments of immunoglobulin molecules, i.e., molecules that contain an antigen binding site.
  • Immunoglobulin molecules can be of any type (e.g., IgG, IgE, IgM, IgD, IgA and IgY), class (e.g., IgG 1 , IgG 2 , IgG 3 , IgG4, IgA 1 and IgA 2 ) or subclass.
  • type e.g., IgG, IgE, IgM, IgD, IgA and IgY
  • class e.g., IgG 1 , IgG 2 , IgG 3 , IgG4, IgA 1 and IgA 2
  • a typical antibody contains two heavy chains paired with two light chains.
  • a full-length heavy chain is about 50 kD in size (approximately 446 amino acids in length), and is encoded by a heavy chain variable region gene (about 116 amino acids) and a constant region gene.
  • There are different constant region genes encoding heavy chain constant region of different isotypes such as alpha, gamma (IgG1, IgG2, IgG3, IgG4), delta, epsilon, and mu sequences.
  • a full-length light chain is about 25 Kd in size (approximately 214 amino acids in length), and is encoded by a light chain variable region gene (about 110 amino acids) and a kappa or lambda constant region gene.
  • the variable regions of the light and/or heavy chain are responsible for binding to an antigen, and the constant regions are responsible for the effector functions typical of an antibody.
  • CDR refers to the complement determining region within antibody variable sequences. There are three CDRs in each of the variable regions of the heavy chain and the light chain, which are designated CDR1, CDR2 and CDR3, for each of the variable regions. The exact boundaries of these CDRs have been defined differently according to different systems.
  • the system described by Kabat Kabat (Kabat et al., Sequences of Proteins of Immunological Interest, National Institutes of Health, Bethesda, Md. (1987) and (1991)) not only provides an unambiguous residue numbering system applicable to any variable region of an antibody, but also provides precise residue boundaries defining the three CDRs. These CDRs may be referred to as Kabat CDRs.
  • Chothia and coworkers found that certain sub-portions within Kabat CDRs adopt nearly identical peptide backbone conformations, despite having great diversity at the level of amino acid sequence. These sub-portions were designated as L1, L2 and L3 or H1, H2 and H3 where the “L” and the “H” designates the light chain and the heavy chains regions, respectively. These regions can be referred to as Chothia CDRs, which have boundaries that overlap with Kabat CDRs.
  • epitopes refers to fragments of a polypeptide or protein having antigenic or immunogenic activity in an animal, preferably in a mammal, and most preferably in a human.
  • An epitope having immunogenic activity is a fragment of a polypeptide or protein that elicits an antibody response in an animal.
  • An epitope having antigenic activity is a fragment of a polypeptide or protein to which an antibody immunospecifically binds as determined by any method well-known to one of skill in the art, for example by immunoassays.
  • Antigenic epitopes need not necessarily be immunogenic.
  • fragment refers to a peptide or polypeptide
  • a fragment of a protein or polypeptide retains at least one function of the protein or polypeptide.
  • immunospecifically binds to an antigen refers to peptides, polypeptides, proteins (including, but not limited to fusion proteins and antibodies) or fragments thereof that specifically bind to an antigen or a fragment and do not specifically bind to other antigens.
  • a peptide, polypeptide, or protein that immunospecifically binds to an antigen may bind to other antigens with lower affinity as determined by, e.g., immunoassays, BIAcore, or other assays known in the art.
  • Antibodies or fragments that immunospecifically bind to an antigen may be cross-reactive with related antigens. Preferably, antibodies or fragments that immunospecifically bind to an antigen do not cross-react with other antigens.
  • the term “in combination” refers to the use of more than one therapies (e.g., more than one prophylactic agent and/or therapeutic agent).
  • the use of the term “in combination” does not restrict the order in which therapies (e.g., prophylactic and/or therapeutic agents) are administered to a subject.
  • a first therapy (e.g., a first prophylactic or therapeutic agent) can be administered prior to (e.g., 5 minutes, 15 minutes, 30 minutes, 45 minutes, 1 hour, 2 hours, 4 hours, 6 hours, 12 hours, 24 hours, 48 hours, 72 hours, 96 hours, 1 week, 2 weeks, 3 weeks, 4 weeks, 5 weeks, 6 weeks, 8 weeks, or 12 weeks before), concomitantly with, or subsequent to (e.g., 5 minutes, 15 minutes, 30 minutes, 45 minutes, 1 hour, 2 hours, 4 hours, 6 hours, 12 hours, 24 hours, 48 hours, 72 hours, 96 hours, 1 week, 2 weeks, 3 weeks, 4 weeks, 5 weeks, 6 weeks, 8 weeks, or 12 weeks after) the administration of a second therapy (e.g., a second prophylactic or therapeutic agent) to a subject.
  • a second therapy e.g., a second prophylactic or therapeutic agent
  • the term “pharmaceutically acceptable” refers approved by a regulatory agency of the federal or a state government, or listed in the U.S. Pharmacopeia, European Pharmacopeia, or other generally recognized pharmacopeia for use in animals, and more particularly, in humans.
  • the terms “prevent,” “preventing,” and “prevention” refer to the inhibition of the development or onset of a disorder or the prevention of the recurrence, onset, or development of one or more symptoms of a disorder in a subject resulting from the administration of a therapy (e.g., a prophylactic or therapeutic agent), or the administration of a combination of therapies (e.g., a combination of prophylactic or therapeutic agents).
  • a therapy e.g., a prophylactic or therapeutic agent
  • a combination of therapies e.g., a combination of prophylactic or therapeutic agents
  • prophylactic agent and “prophylactic agents” refer to any agent(s) which can be used in the prevention of a disorder or one or more of the symptoms thereof.
  • the term “prophylactic agent” refers to an antibody of the invention.
  • the term “prophylactic agent” refers to an agent other than an antibody of the invention.
  • a prophylactic agent is an agent which is known to be useful to or has been or is currently being used to the prevent or impede the onset, development, progression and/or severity of a disorder or one or more symptoms thereof.
  • prophylactically effective amount refers to the amount of a therapy (e.g., prophylactic agent) which is sufficient to result in the prevention of the development, recurrence, or onset of a disorder or one or more symptoms thereof, or to enhance or improve the prophylactic effect(s) of another therapy (e.g., a prophylactic agent).
  • a therapy e.g., prophylactic agent
  • the terms “subject” and “patient” are used interchangeably.
  • the terms “subject” and “subjects” refer to an animal, preferably a mammal including a non-primate (e.g., a cow, pig, horse, cat, dog, rat, and mouse) and a primate (e.g., a monkey, such as a cynomolgous monkey, a chimpanzee, and a human), and most preferably a human.
  • a non-primate e.g., a cow, pig, horse, cat, dog, rat, and mouse
  • a primate e.g., a monkey, such as a cynomolgous monkey, a chimpanzee, and a human
  • the subject is a non-human animal such as a bird (e.g., a quail, chicken, or turkey), a farm animal (e.g., a cow, horse, pig, or sheep), a pet (e.g., a cat, dog, or guinea pig), or laboratory animal (e.g., an animal model for a disorder).
  • a bird e.g., a quail, chicken, or turkey
  • a farm animal e.g., a cow, horse, pig, or sheep
  • a pet e.g., a cat, dog, or guinea pig
  • laboratory animal e.g., an animal model for a disorder
  • the subject is a human (e.g., an infant, child, adult, or senior citizen).
  • a therapeutic agent refers to any agent(s) which can be used in the prevention, treatment, management, or amelioration of a disorder or one or more symptoms thereof.
  • the term “therapeutic agent” refers to an antibody of the invention.
  • the term “therapeutic agent” refers an agent other than an antibody of the invention.
  • a therapeutic agent is an agent which is known to be useful for, or has been or is currently being used for the prevention, treatment, management, or amelioration of a disorder or one or more symptoms thereof.
  • the term “therapeutically effective amount” refers to the amount of a therapy (e.g. an antibody of the invention), which is sufficient to reduce the severity of a disorder, reduce the duration of a disorder, ameliorate one or more symptoms of a disorder, prevent the advancement of a disorder, cause regression of a disorder, or enhance or improve the therapeutic effect(s) of another therapy.
  • a therapy e.g. an antibody of the invention
  • the terms “therapies” and “therapy” can refer to any protocol(s), method(s), and/or agent(s) that can be used in the prevention, treatment, management, and/or amelioration of a disorder or one or more symptoms thereof.
  • the terms “therapy” and “therapy” refer to anti-viral therapy, anti-bacterial therapy, anti-fungal therapy, anti-cancer agent, biological therapy, supportive therapy, and/or other therapies useful in treatment, management, prevention, or amelioration of a disorder or one or more symptoms thereof known to one skilled in the art, for example, a medical professional such as a physician.
  • the terms “treat,” “treatment,” and “treating” refer to the reduction or amelioration of the progression, severity, and/or duration of a disorder or amelioration of one or more symptoms thereof resulting from the administration of one or more therapies (including, but not limited to, the administration of one or more prophylactic or therapeutic agents).
  • sequence alignment refers to the result when at least two antibody sequences are compared for maximum correspondence, as measured using a sequence comparison algorithms.
  • Optimal alignment of sequences for comparison can be conducted by any technique known or developed in the art, and the invention is not intended to be limited in the alignment technique used.
  • Exemplary alignment methods include the local homology algorithm of Smith & Waterman, Adv. Appl. Math. 2:482 (1981), the homology alignment algorithm of Needleman & Wunsch, J. Mol. Biol. 48:443 (1970), the search for similarity method of Pearson & Lipman, Proc. Nat'l. Acad. Sci.
  • the “three dimensional structure” of a protein is also termed the “tertiary structure” or the structure of the protein in three dimensional space.
  • the three dimensional structure of a protein is determined through X-ray crystallography and the coordinates of the atoms of the amino acids determined. The coordinates are then converted through an algorithm into a visual representation of the protein in three dimensional space. From this model, the local “environment” of each residue can be determined and the “solvent accessibility” or exposure of a residue to the extraprotein space can be determined.
  • the “proximity of a residue to a site of functionality” or active site and more specifically, the “distance of the ⁇ or ⁇ carbons of the residue to the site of functionality” can be determined.
  • the ⁇ carbon can be substituted.
  • residues that “contact with residues of interest” can be determined. These would be residues that are close in three dimensional space and would be expected to form bonds or interactions with the residues of interest. And because of the electron interactions across bonds, residues that contact residues in contact with residues of interest can be investigated for possible mutability.
  • nuclear magnetic resonance spectroscopy can be used to determine the structure.
  • molecular modeling can be used to determine the structure, and can be based on an homologous structure or ab initio. Energy minimization techniques can also be employed.
  • Residue chemistry refers to characteristics that a residue possesses in the context of a protein or by itself. These characteristics include, but are not limited to, polarity, hydrophobicity, net charge, molecular weight, propensity to form a particular secondary structure, and space filling size.
  • carrier refers to a diluent, adjuvant, excipient, or vehicle.
  • Carriers can be liquids, such as water and oils, including those of petroleum, animal, vegetable or synthetic origin, such as peanut oil, soybean oil, mineral oil, sesame oil and the like.
  • the vehicles e.g., pharmaceutical vehicles
  • the vehicles can be saline, gum acacia, gelatin, starch paste, talc, keratin, colloidal silica, urea, and the like.
  • auxiliary, stabilizing, thickening, lubricating and coloring agents can be used.
  • the carriers When administered to a patient, the carriers are preferably sterile. Water can be the carrier when composition is administered intravenously.
  • Saline solutions and aqueous dextrose and glycerol solutions can also be employed as liquid vehicles, particularly for injectable solutions.
  • Suitable vehicles also include excipients such as starch, glucose, lactose, sucrose, gelatin, malt, rice, flour, chalk, silica gel, sodium stearate, glycerol monostearate, talc, sodium chloride, dried skim milk, glycerol, propyleneglycol, water, ethanol and the like.
  • Compositions if desired, can also contain minor amounts of wetting or emulsifying agents, or pH buffering agents.
  • the term “functional domain” means a segment of a protein that has one or more of the following properties (i) a structurally independent section of a protein, (ii) a section of a protein that is homologous to a section of another protein, (iii) a segment of protein involved in one or more specific functions, (iv) an independently evolving unit in a protein, (v) a segment of protein containing a particular sequence motif, (vi) a section of the protein containing an active site, a binding site or a regulatory site.
  • An expert system 100 is computer program that represents and reasons with the knowledge of some specialist subject (antibodies) with a view to solving problems or giving advice (via rank ordering of substitutions with reasoning)
  • Knowledge acquisition is the transfer and transformation of potential problem-solving expertise (e.g. knowledge of analysing nucleotide or protein structure, nucleotide or protein phylogeny) from the knowledge source to a program.
  • potential problem-solving expertise e.g. knowledge of analysing nucleotide or protein structure, nucleotide or protein phylogeny
  • Knowledge base 108 is the encoded knowledge for an expert system 100 .
  • a knowledge base 108 typically incorporates definitions of attributes and rules along with control information.
  • An inference engine 106 is software that provides the reasoning mechanism in expert system 100 .
  • it typically implements forward chaining and backward chaining strategies.
  • a substitution in an antibody is the replacement of one monomer with a different monomer.
  • a virtual surrogate screen is a measure of the activity of an antibody in dimensions that are mathematically constructed from physical measurements of antibody properties in two or more assays.
  • screen, assay, test and measurement are used interchangeably to mean a method of determining one or more property of an antibody.
  • a high throughput screen, assay, test or measurement is used to describe any method for determining one or more property of a plurality of antibodies either sequentially of simultaneously.
  • the actual number of antibody variants whose properties can be determined by a test that is considered a high throughput screen varies from as few as 84 samples per day (Decker et al. (2003) Appl Biochem Biotechnol 105: 689-703.) to many millions.
  • a high throughput screen as an assay that can measure one or more antibody property for 400 antibody variants in 1 week, preferably a test that can measure one or more antibody property for 1,000 antibody variants in 1 week, more preferably a test that can measure one or more antibody properties for 10,000 antibody variants in one week.
  • Antibody variants can be synthesized by methods for constructing or obtaining specific nucleic acid or polypeptide sequences described in the art. Antibody variants are designed, for example, in step 03 of FIG. 2 , as described in Section 5.2, above.
  • Oligonucleotides and polynucleotides can be synthesized using a variety of chemistries including phosphoramidite chemistry; optionally this synthesis may be performed using a commercially available DNA synthesizer. Oligonucleotides and polynucleotides may also be purchased from a commercial supplier of synthetic DNA.
  • Chemically synthesized oligonucleotides can be incorporated into larger polynucleotides to create one or more of the designed sequence variants using site-directed mutagenesis.
  • Suitable site-directed techniques include those in which a template strand is used to prime the synthesis of a complementary strand lacking a modification in the parent strand, such as methylation or incorporation of uracil residues; introduction of the resulting hybrid molecules into a suitable host strain results in degradation of the template strand and replication of the desired mutated strand. See (Kunkel (1985) Proc Natl Acad Sci USA 82: 488-92.); QuikChange kits available from Stratagene, Inc., La Jolla, Calif.
  • PCR methods for introducing site-directed changes can also be employed.
  • Site-directed mutagenesis using a single stranded DNA template and mutagenic oligos is well known in the art (Ling & Robinson 1997, Anal Biochem 254:157 1997). It has also been shown that several oligos can be incorporated at the same time using these methods (Zoller 1992, Curr Opin Biotechnol 3: 348).
  • Single stranded DNA templates are synthesized by degrading double stranded DNA (StrandaseTM by Novagen). The resulting product after strain digestion can be heated and then directly used for sequencing.
  • the template can be constructed as a phagemid or M13 vector.
  • Other techniques of incorporating mutations into DNA are known and can be found in, e.g., Deng et al. 1992, Anal Biochem 200:81.
  • Oligonucleotides can together be assembled into larger polynucleotides to create one or more of the designed sequence variants.
  • Oligonucleotides can be assembled into larger single- or double-stranded polynucleotides in vivo or in vitro by a variety of methods including but not limited to annealing, restriction enzyme digestion and ligation, particularly using restriction enzymes whose cleavage site is distinct from their recognition sites (see for example Pierce 1994, Biotechniques 16:708-15; Mandecki & Bolling 1988, Gene 68:101-7), ligation (see for example Edge at al 1981, Nature 292:756-62; Jayaraman & Puccini 1992 Biotechniques 12:392-8), ligation followed by polymerase chain reaction amplification (see for example Jayaraman et al 1991, Proc Natl Acad Sci USA.
  • thermostable nucleotide polymerases and/or ligases see for example Ye et al. 1992, Biochem Biophys Res Commun. 186:143-9; Horton et al 1989 Gene. 77:61-8; Stemmer et al 1995 Gene. 164:49-53
  • dual asymmetric PCR see for example Sandhu et al 1992, Biotechniques 12:14-6) stepwise elongation of sequences (see for example Majumder 1992, Gene. 110:89-94), the ligase chain reaction (see for example Au et al 1998, Biochem Biophys Res Commun.
  • oligonucleotides sequential ligation of one or more oligonucleotides to an anchored oligonucleotide (for example a biotinylated oligonucleotide immobilized on streptavidin resin), cotransformation into an appropriate host cell such as mammalian, yeast or bacterial cells capable of joining polynucleotides (see for example Raymond et al 1999, BioTechniques 26: 134-141), or any combination of steps involving the activity of one or more of a polymerase, a ligase, a restriction enzyme, and a recombinase.
  • an appropriate host cell such as mammalian, yeast or bacterial cells capable of joining polynucleotides
  • Oligonucleotides can optionally be designed to improve their assembly into larger polynucleotides and subsequent processing, for example by optimizing annealing properties and eliminating restriction sites (see for example Hoover & Lubkowski 2002, Nucleic Acids Res. 30:e43).
  • Synthesis of polynucleotide sequence variants can also be multiplexed. Individual variants can subsequently be identified, for example by picking and sequencing single clones. Other methods of deconvolution include testing for an easily measured phenotype (examples include but are not limited to colorigenic, fluorigenic or turbidity-altering reactions that can be visualized on agar plates), then grouping clones according to activity and selecting one or more clone from each group. Optionally the one or more clone from each group may be sequenced.
  • multiplexed variant synthesis is to incorporate one or more oligonucleotides containing one or more alternative nucleotide substitutions into one or more polynucleotide reference sequences simultaneously.
  • Oligonucleotides synthesized from mixtures of nucleotides can be used. The synthesis of oligonucleotide libraries is well known in the art.
  • degenerate oligos from trinucleotides can be used (Gaytan, et al., 1998, Chem Biol 5:519; Lyttle, et al 1995, Biotechniques 19:274; Virnekas, et al 1994, Nucl.
  • degenerate oligos can be synthesized by resin splitting (Lahr, et al 1999, Proc. Natl. Acad. Sci. USA 96:14860; Chatellier, et al., 1995, Anal. Biochem. 229:282; and Haaparanta & Huse 1995, Mol Divers 1:39). Mixtures of individual primers for the substitutions to be introduced by site directed mutagenesis can be simultaneously employed in a single reaction to produce the desired combinations of mutations.
  • Simultaneous mutation of adjacent residues can be accomplished by preparing a plurality of oligonucleotides representing the desired combinations.
  • sequences are assembled using PCR to link synthetic oligos (Horton, et al 1989, Gene 77:61; Shi, et al 1993, PCR Methods Appl. 3:46; and Cao 1990, Technique 2:109).
  • PCR with a mixture of mutagenic oligos can be used to create a multiplexed set of sequence variants that can subsequently be deconvoluted.
  • Cassette mutagenesis can also be used in creating multiple polynucleotide sequence variants.
  • a set of sequences can be generated by ligating fragments obtained by oligonucleotide synthesis, PCR or combinations thereof. Segments for ligation can, for example, be generated by PCR and subsequent digestion with type II restriction enzymes. This enables introduction of mutations via the PCR primers. Furthermore, type II restriction enzymes generate non-palindromic cohesive ends which significantly reduce the likelihood of ligating fragments in the wrong order. Techniques for ligating many fragments can be found in Berger, et al., Anal Biochem 214:571 (1993).
  • Antibody variants can be synthesized as nucleic acid sequence variants according to any of the processes described here, followed by expression either in vivo or in an in vitro cell-free system. They may also be made directly using commercial peptide synthesizers. Antibody variants may additionally be synthesized by chemically ligating one or more synthetic peptides to one or more polypeptide segments created by expression of a polynucleotide (see for example Pal et al 2003 Protein Expr Purif. 29:185-92).
  • Antibody variants may optionally include non-natural amino acids, incorporated at specific positions in the protein sequence by a variety of methods (see for example Hyun Bae et al 2003, J Mol. Biol. 328:1071-81; Hohsaka & Sisido 2002, Curr Opin Chem. Biol. 6:809-15; L1 and Roberts 2003, Chem. Biol 10:233-9).
  • the particular chemical and/or molecular biological methods used to construct the antibody sequence variants are not critical; any method(s) that provide the desired sequence variants can be used.
  • Section 5.2 described how a designed set of antibody variants was designed. This set of antibodies is then synthesized using, for example, the techniques described in Section 5.8. Then the antibodies are tested for relevant biological activity and/or antibody properties. Determination of what constitutes a relevant antibody property is a case specific exercise.
  • Non-limiting examples of antibody properties that can be relevant in some embodiments of the present invention include, but are not limited, to antigenicity, immunogenicity, immunomodulatory activity, expression of the antibody in a homologous host, expression of the antibody in a heterologous host, expression of the antibody in a plant cell, susceptibility of the antibody to in vitro post-translational modifications and susceptibility of the antibody to in vivo post-translational modifications.
  • High throughput screens typically do not measure the complex combination of functions that are desired in the final engineered antibody.
  • High throughput screens can be used to measure some properties of the antibody, and the method of this invention allows the properties measured in two or more of these high throughput screens to be combined and used to create a virtual surrogate screen for the properties of interest.
  • High throughput screens that may be used to measure potentially relevant antibody properties include but are not limited to: flow cytometry (Daugherty et al. (2000) J Immunol Methods 243: 211-27.; Georgiou (2000) Adv Protein Chem 55: 293-315.; Olsen et al. (2000) Curr Opin Biotechnol 11: 331-7.) solid phase digital imaging (Joo et al.
  • Measurements of cell lines and primary cell cultures for cell-surface receptor surface density measurements of cell surface receptor internalization rates, cell surface receptor post-translational modifications including phosphorylation, binding of antigens including but not limited to cellular growth factor receptors, receptors or mediators of tumor-driven angiogenesis, B cell surface antigens and proteins synthesized by or in response to pathogens, antigens produced by the induction of antibody-mediated cell killing, antigens produced by antibody-dependent macrophage activity, histamine, and antigens produced by induction of or cross-reaction with anti-idiotype antibodies.
  • Examples of antibody properties or activities whose measurement may be resource, time or cost-limited and that therefore cannot be accurately measured in high throughput are tests for the immunogenicity of an antibody, in vivo or cell-culture based viral titer measurements, any experiment in which an experimental animal or human being is used as a part of the measurement of one or more properties of the antibody, the level of expression of the antibody in a host, any experiment in which the antibody is produced within a plant particularly when the plant must be transformed with a polynucleotide encoding the antibody and the antibody be expressed within the plant, susceptibility of the antibody to be modified inside a living cell, susceptibility of the antibody to be modified not inside a living cell, measurement of the composition of a complex mixture of compounds whose composition has been altered by the action of the antibody (for example metabolomics or metabonomics, alteration of the properties of a cell for example alteration of the growth, replication or differentiation patterns of a cell or population of cells, therapeutic efficacy of an antibody and modulation of a signaling pathway.
  • Antibodies of the present invention or fragments thereof can be assayed in a variety of ways well-known to one of skill in the art.
  • antibodies of the invention or fragments thereof can be assayed for the ability to immunospecifically bind to an antigen.
  • Such an assay can be performed in solution (e.g., Houghten, 1992, Bio/Techniques 13:412 421), on beads (Lam, 1991, Nature 354:82 84), on chips (Fodor, 1993, Nature 364:555 556), in bacteria (U.S. Pat. No. 5,223,409), in spores (U.S. Pat. Nos.
  • Immunoassays that can be used to analyze immunospecific binding and cross-reactivity include, but are not limited to, competitive and non-competitive assay systems using techniques such as Western blots, radioimmunoassays, ELISA (enzyme linked immunosorbent assay), “sandwich” immunoassays, immunoprecipitation assays, precipitin reactions, gel diffusion precipitin reactions, immunodiffusion assays, agglutination assays, complement-fixation assays, immunoradiometric assays, fluorescent immunoassays, protein A immunoassays, to name but a few.
  • Antibodies of the invention or fragments thereof can also be assayed for their ability to inhibit the binding of an antigen to its host cell receptor using techniques known to those of skill in the art. For example, cells expressing a receptor can be contacted with a ligand for that receptor in the presence or absence of an antibody or fragment thereof that is an antagonist of the ligand and the ability of the antibody or fragment thereof to inhibit the ligand's binding can measured by, for example, flow cytometry or a scintillation assay.
  • the ligand or the antibody or antibody fragment can be labeled with a detectable compound such as a radioactive label (e.g., 32 P, 35 S, and 125 I) or a fluorescent label (e.g., fluorescein isothiocyanate, rhodamine, phycoerythrin, phycocyanin, allophycocyanin, o-phthaldehyde and fluorescamine) to enable detection of an interaction between the ligand and its receptor.
  • a detectable compound such as a radioactive label (e.g., 32 P, 35 S, and 125 I) or a fluorescent label (e.g., fluorescein isothiocyanate, rhodamine, phycoerythrin, phycocyanin, allophycocyanin, o-phthaldehyde and fluorescamine) to enable detection of an interaction between the ligand and its receptor.
  • a detectable compound such as a radioactive label (e.g.,
  • a ligand can be contacted with an antibody or fragment thereof that is an antagonist of the ligand and the ability of the antibody or antibody fragment to inhibit the ligand from binding to its receptor can be determined.
  • the antibody or the antibody fragment that is an antagonist of the ligand is immobilized on a solid support and the ligand is labeled with a detectable compound.
  • the ligand is immobilized on a solid support and the antibody or fragment thereof is labeled with a detectable compound.
  • a ligand can be partially or completely purified (e.g., partially or completely free of other polypeptides) or part of a cell lysate.
  • a ligand can be biotinylated using techniques well known to those of skill in the art (e.g., biotinylation kit, Pierce Chemicals; Rockford, Ill.).
  • An antibody or a fragment thereof constructed and/or identified in accordance with the present invention can be tested in vitro and/or in vivo for its ability to modulate the biological activity of cells. Such ability can be assessed by, e.g., detecting the expression of antigens and genes; detecting the proliferation of cells; detecting the activation of signaling molecules (e.g., signal transduction factors and kinases); detecting the effector function of cells; or detecting the differentiation of cells. Techniques known to those of skill in the art can be used for measuring these activities. For example, cellular proliferation can be assayed by 3 H-thymidine incorporation assays and trypan blue cell counts.
  • Antigen expression can be assayed, for example, by immunoassays including, but are not limited to, competitive and non-competitive assay systems using techniques such as western blots, immunohistochemistry radioimmunoassays, ELISA (enzyme linked immunosorbent assay), “sandwich” immunoassays, immunoprecipitation assays, precipitin reactions, gel diffusion precipitin reactions, immunodiffusion assays, agglutination assays, complement-fixation assays, immunoradiometric assays, fluorescent immunoassays, protein A immunoassays, and FACS analysis.
  • the activation of signaling molecules can be assayed, for example, by kinase assays and electrophoretic shift assays (EMSAs).
  • the antibodies, fragments thereof, or compositions of the invention are preferably tested in vitro and then in vivo for the desired therapeutic or prophylactic activity prior to use in humans.
  • assays which can be used to determine whether administration of a specific pharmaceutical composition is indicated include cell culture assays in which a patient tissue sample is grown in culture and exposed to, or otherwise contacted with, a pharmaceutical composition, and the effect of such composition upon the tissue sample is observed.
  • the tissue sample can be obtained by biopsy from the patient. This test allows the identification of the therapeutically most effective therapy (e.g., prophylactic or therapeutic agent) for each individual patient.
  • in vitro assays can be carried out with representative cells of cell types involved a particular disorder to determine if a pharmaceutical composition of the invention has a desired effect upon such cell types.
  • in vitro assay can be carried out with cell lines.
  • Peripheral blood lymphocytes counts in a subject can be determined by, e.g., obtaining a sample of peripheral blood from said subject, separating the lymphocytes from other components of peripheral blood such as plasma using, e.g., Ficoll-Hypaque (Pharmacia) gradient centrifugation, and counting the lymphocytes using trypan blue.
  • Peripheral blood lymphocytes counts in a subject can be determined by, e.g., obtaining a sample of peripheral blood from said subject, separating the lymphocytes from other components of peripheral blood such as plasma using, e.g., Ficoll-Hypaque (Pharmacia) gradient centrifugation, and counting the lymphocytes using trypan blue.
  • Peripheral blood T-cell counts in subject can be determined by, e.g., separating the lymphocytes from other components of peripheral blood such as plasma using, e.g., a use of Ficoll-Hypaque (Pharmacia) gradient centrifugation, labeling the T-cells with an antibody directed to a T-cell antigen which is conjugated to FITC or phycoerythrin, and measuring the number of T-cells by FACS.
  • Ficoll-Hypaque Pharmacia
  • the antibodies, fragments, or compositions of the invention used to treat, manage, prevent, or ameliorate a viral infection or one or more symptoms thereof can be tested for their ability to inhibit viral replication or reduce viral load in in vitro assays.
  • viral replication can be assayed by a plaque assay such as described, e.g., by Johnson et al., 1997, Journal of Infectious Diseases 176:1215-1224 176:1215-1224.
  • the antibodies or fragments thereof administered according to the methods of the invention can also be assayed for their ability to inhibit or downregulate the expression of viral polypeptides. Techniques known to those of skill in the art, including, but not limited to, western blot analysis, northern blot analysis, and RT-PCR can be used to measure the expression of viral polypeptides.
  • Antibodies, fragments, or compositions of the invention can be tested in additional in vitro assays that are well-known in the art.
  • additional In vitro assays known in the art can also be used to test the existence or development of resistance of bacteria to a therapy.
  • Such in vitro assays are described in Gales et al., 2002, Diag. Nicrobiol. Infect. Dis. 44(3):301-311; Hicks et al., 2002, Clin. Microbiol. Infect. 8(11): 753-757; and Nicholson et al., 2002, Diagn. Microbiol. Infect. Dis. 44(1): 101-107.
  • the antibodies, fragments, or compositions of the invention can be assayed for the ability to treat, manage, prevent, or ameliorate a fungal infection or one or more symptoms thereof.
  • Any of the standard anti-fungal assays well-known in the art can be used to assess such activity. For instance, tests recommended by the National Committee for Clinical Laboratories (NCCLS) (See National Committee for Clinical Laboratories Standards. 1995, Proposed Standard M27T. Villanova, Pa., all of which is incorporated herein by reference in its entirety) and other methods known to those skilled in the art (Pfaller et al., 1993, Infectious Dis. Clin. N. Am. 7: 435-444) can be used.
  • Such antifungal properties can also be determined from a fungal lysis assay, as well as by other methods, including, inter alia, growth inhibition assays, fluorescence-based fungal viability assays, flow cytometry analyses, and other standard assays known to those skilled in the art.
  • any in vitro assays known to those skilled in the art can be used to evaluate the prophylactic and/or therapeutic utility of an antibody disclosed herein for a particular disorder or one or more symptoms thereof.
  • the antibodies, compositions, or combination therapies of the invention can be tested in suitable animal model systems prior to use in humans.
  • animal model systems include, but are not limited to, rats, mice, chicken, cows, monkeys, pigs, dogs, rabbits, etc. Any animal system well-known in the art may be used.
  • Several aspects of the procedure may vary; such aspects include, but are not limited to, the temporal regime of administering the therapies (e.g., prophylactic and/or therapeutic agents) whether such therapies are administered separately or as an admixture, and the frequency of administration of the therapies.
  • Animal models can be used to assess the efficacy of the antibodies, fragments thereof, or compositions of the invention for treating, managing, preventing, or ameliorating a particular disorder or one or more symptom thereof.
  • the antibodies, fragments thereof of compositions of the present invention can be assayed for their ability to decrease the time course of a particular disorder by at least 25%, preferably at least 50%, at least 60%, at least 75%, at least 85%, at least 95%, or at least 99%.
  • the antibodies, compositions, or combination therapies of the invention can also be assayed for their ability to increase the survival period of organisms (e.g., humans) suffering from a particular disorder by at least 25%, preferably at least 50%, at least 60%, at least 75%, at least 35%, at least 95%, or at least 99%.
  • antibodies, fragments thereof, compositions, or combination therapies of the invention can be assayed their ability reduce the hospitalization period of humans suffering from viral respiratory infection by at least 60%, preferably at least 75%, at least 35%, at least 95%, or at least 99%.
  • the toxicity and/or efficacy of the antibodies, fragments thereof, or compositions of the present invention can be assayed by standard pharmaceutical procedures in cell cultures or experimental animals, e.g., for determining the LD50 (the dose lethal to 50% of the population) and the ED50 (the dose therapeutically effective in 50% of the population).
  • the dose ratio between toxic and therapeutic effects is the therapeutic index and it can be expressed as the ratio LD50/ED50.
  • Antibodies that exhibit large therapeutic indices are preferred. While antibodies that exhibit toxic side effects can be used, care should be taken to design a delivery system that targets such agents to the site of affected tissue in order to minimize potential damage to uninfected cells and, thereby, reduce side effects.
  • kits comprising a set of variant or a single variant in a set of variants that have been refined by the apparatus and methods describe herein.
  • the invention also provides a pharmaceutical pack or kit comprising one or more containers filled with a variant of set of variants of the present invention.
  • the pharmaceutical pack or kit may further comprise one or more other prophylactic or therapeutic agents useful for the treatment of a particular disease.
  • the invention also provides a pharmaceutical pack or kit comprising one or more containers filled with one or more of the ingredients of the pharmaceutical compositions of the invention.
  • Optionally associated with such container(s) can be a notice in the form prescribed by a governmental agency regulating the manufacture, use or sale of pharmaceuticals or biological products, which notice reflects approval by the agency of manufacture, use or sale for human administration.
  • the present invention also encompasses a finished packaged and labeled pharmaceutical product.
  • This article of manufacture includes the appropriate unit dosage form in an appropriate vessel or container such as a glass vial or other container that is hermetically sealed.
  • the active ingredient is sterile and suitable for administration as a particulate free solution.
  • the invention encompasses both parenteral solutions and lyophilized powders, each being sterile, and the latter being suitable for reconstitution prior to injection.
  • the unit dosage form may be a solid suitable for oral, transdermal, topical or mucosal delivery.
  • the unit dosage form is suitable for intravenous, intramuscular or subcutaneous delivery.
  • the invention encompasses solutions, preferably sterile, suitable for each delivery route.
  • the packaging material and container are designed to protect the stability of the product during storage and shipment.
  • the products of the invention include instructions for use or other informational material that advise the physician, technician or patient on how to appropriately prevent or treat the disease or disorder in question.
  • the article of manufacture includes instruction means indicating or suggesting a dosing regimen including, but not limited to, actual doses, monitoring procedures (such as methods for monitoring mean absolute lymphocyte counts, tumor cell counts, and tumor size) and other monitoring information.
  • the invention provides an article of manufacture comprising packaging material, such as a box, bottle, tube, vial, container, sprayer, insufflator, intravenous (i.v.) bag, envelope and the like; and at least one unit dosage form of a pharmaceutical agent contained within said packaging material.
  • packaging material such as a box, bottle, tube, vial, container, sprayer, insufflator, intravenous (i.v.) bag, envelope and the like
  • at least one unit dosage form of a pharmaceutical agent contained within said packaging material such as a box, bottle, tube, vial, container, sprayer, insufflator, intravenous (i.v.) bag, envelope and the like; and at least one unit dosage form of each pharmaceutical agent contained within said packaging material.
  • an article of manufacture comprises packaging material and a pharmaceutical agent and instructions contained within said packaging material, wherein said pharmaceutical agent is a humanized antibody and a pharmaceutically acceptable carrier, and said instructions indicate a dosing regimen for preventing, treating or managing a subject with a particular disease.
  • an article of manufacture comprises packaging material and a pharmaceutical agent and instructions contained within said packaging material, wherein said pharmaceutical agent is a humanized antibody, a prophylactic or therapeutic agent other than the humanized antibody and a pharmaceutically acceptable carrier, and said instructions indicate a dosing regimen for preventing, treating or managing a subject with a particular disease.
  • an article of manufacture comprises packaging material and two pharmaceutical agents and instructions contained within said packaging material, wherein said first pharmaceutical agent is a humanized antibody and a pharmaceutically acceptable carrier and said second pharmaceutical agent is a prophylactic or therapeutic agent other than the humanized antibody, and said instructions indicate a dosing regimen for preventing, treating or managing a subject with a particular disease.
  • the present invention provides that the adverse effects that may be reduced or avoided by the methods of the invention are indicated in informational material enclosed in an article of manufacture for use in preventing, treating or ameliorating one or more symptoms associated with a disease.
  • Adverse effects that may be reduced or avoided by the methods of the invention include but are not limited to vital sign abnormalities (e.g., fever, tachycardia, bardycardia, hypertension, hypotension), hematological events (e.g., anemia, lymphopenia, leukopenia, thrombocytopenia), headache, chills, dizziness, nausea, asthenia, back pain, chest pain (e.g., chest pressure), diarrhea, myalgia, pain, pruritus, psoriasis, rhinitis, sweating, injection site reaction, and vasodilatation. Since some of the therapies may be immunosuppressive, prolonged immunosuppression may increase the risk of infection, including opportunistic infections. Prolonged and sustained immunosuppression may
  • the information material enclosed in an article of manufacture can indicate that foreign proteins may also result in allergic reactions, including anaphylaxis, or cytosine release syndrome.
  • the information material should indicate that allergic reactions may exhibit only as mild pruritic rashes or they may be severe such as erythroderma, Stevens Johnson syndrome, vasculitis, or anaphylaxis.
  • the information material should also indicate that anaphylactic reactions (anaphylaxis) are serious and occasionally fatal hypersensitivity reactions.
  • Allergic reactions including anaphylaxis may occur when any foreign protein is injected into the body. They may range from mild manifestations such as urticaria or rash to lethal systemic reactions. Anaphylactic reactions occur soon after exposure, usually within 10 minutes.
  • Patients may experience paresthesia, hypotension, laryngeal edema, mental status changes, facial or pharyngeal angioedema, airway obstruction, bronchospasm, urticaria and pruritus, serum sickness, arthritis, allergic nephritis, glomerulonephritis, temporal arthritis, or eosinophilia.
  • sequence variants of proteinase K is described here as an example of the use of sequence-activity relationships to engineer desired properties into a protein. Also described is the analysis of these variants using six different functional tests, and methods for determining components of a virtual screen.
  • FIG. 6 shows the amino acid sequence of proteinase K that occurs naturally in the fungus Tritirachium album Limber (Gunkel et al. (1989) Eur J Biochem 179: 185-194) (SEQ ID NO.: 2) together with an E. coli leader peptide (SEQ ID NO.: 1).
  • FIG. 7 shows a nucleotide sequence designed to encode proteinase K (SEQ ID NO.: 3). The sequence has been modified from the original Tritirachium album sequence by removing an intron, adding an E. coli leader peptide and altering the codons used to resemble the distribution found in the highly expressed genes of E. coli . The gene was synthesized for the natural proteinase K from oligonucleotides.
  • the proteinase K gene was used as probe against GenBank using BLAST based algorithms.
  • a BLAST score was chosen as a cut-off that identified more than ten but less than one hundred related sequences. This search identified the 49 sequences identified in FIG. 8 .
  • the sequences (49 rows ⁇ 728 variables) were represented in a Free-Wilson method of qualitative binary description of monomers (Kubinyi, 3D QSAR in drug design theory methods and applications. Pergamon Press, Oxford, 1990, pp 589-638), and distributed in a maximally compressed space using principal component analysis so that the first principal component (PC) captured 10.8 percent of all variance information (eigenvalue of 79), the second principal component (PC) captured 7.8 percent of all variance information (eigenvalue of 57), the third principal component (PC) captured 6.9 percent of all variance information (eigenvalue of 50), the fourth principal component (PC) captured 6.2 percent of all variance information (eigenvalue of 45), the fifth principal component (PC) captured 5.4 percent of all variance information (eigenvalue of 39) and so on until 728 th principal component (PC) captured 0 percent of all variance information (Eigenvalue 0).
  • PC principal component
  • Sequences 46, 47, 48, 49 are all derived from thermophilic organisms and are all well separated from the proteinase K homologs 1-45 in both of the first two principal components, as shown in FIG. 9 .
  • a corresponding plot of all loads describes the influence of each variable on the sample distribution in the various PC's.
  • the correlation between loads (influence of variables—in this case amino acid residues) and score (distribution of samples—here proteinase K homologs) illustrates graphically which residues are unique in determining the phylogenetic separation of genes 46-49 from genes 1-45. This is shown in FIG. 10 .
  • residues that are completely co-evolving will have the exact same load and consequently collapse the variable space in as many dimensions as there are absolute coevolving residues.
  • residues 15D, 18D, 19Q, 22L, 23P, 65Y, 66D, 110R, 137P, 164D, 189C, 198R all are completely co-evolving and all have profound effect on the distribution of samples 46-49 in PC 1 and PC 2 .
  • Variables here can be amino acids as depicted in this example, or any type of feature.
  • Features include, but are not limited to, physico-chemical properties of one or more amino acid residues.
  • the residues can be a block or modulated within the gene, or it can be a combination of residues not genetically linked such as in the example above of residues 15D, 18D, 19Q, 22L, 23P, 65Y, 66D, 110R, 137P, 164D, 189C, 198R.
  • thermophilic proteinase K homologs The loads for the amino acids most responsible for the clustering of thermophilic proteinase K homologs are shown in FIG. 12 . This information was then incorporated into knowledge base 108 . This is an example of pre-processing information.
  • thermostable enzymes known to be thermostable. These enzymes were aligned positions that were conserved between the thermostable homologs but not found in non-thermostable homologs were identified. This information was then incorporated into the knowledge base 108 . This is an example of pre-processing information.
  • thermostable A score for change found in a close homolog that is thermostable but not in close homologs that are not thermostable was added.
  • variants with six changes per clone were designed. In this example all of the 24 top-scoring changes were equally represented. In other embodiments, a set of variants that represent each change with a frequency reflecting its actual score could have been designed. In this case, 24 clones were designed that cover the sequence space uniformly. One way to measure the uniformity of the space covered is by counting the number of instances a particular substitution (e.g., N95C) is seen in the 24 clones. This number was set at six for all the variations identified. This means, that in the set of variations synthesized, each of the identified mutations occurs six times. For example, the mutation N95C is found in six of the variants, the mutation P97S is found in six of the variants, and so forth.
  • N95C is found in six of the variants
  • the mutation P97S is found in six of the variants, and so forth.
  • FIG. 14 The variants defined by this process are listed in FIG. 14 , where FIG. 13 serves as the key for FIG. 14 .
  • “95 in FIG. 14 means “N95C”
  • “355 in FIG. 14 means “P335S”.
  • each proteinase K variant defined in FIG. 14 was synthesized by PCR-based assembly of synthetic oligonucleotides. The sequence of each variant was confirmed using an ABI sequencer. The ability of each of these variant proteins to hydrolyze casein was then measured simply to determine whether the proteinase K variants had any protease activity. This is the first step in exploring the sequence space. ( FIG. 2 , step 04 ).
  • This data can be used to analyze the data using sequence-activity correlating methods to evaluate the substitutions (steps 05 and 06 of FIG. 2 ).
  • this information can be used to update knowledge base 108 and to perform additional iterations of the method to thus further explore the sequence space for improvements in desired properties.
  • expert system 100 in conjunction with the sequence-activity correlating methods inferred that the proline to serine change (seen at positions 97 and 265 ) for flexibility and structural perturbation twice resulted in disadvantageous changes.
  • This information was coded into the knowledge base 108 for future experiments. This is one illustration of updating knowledge base 108 .
  • each constructed variant is shown in FIG. 16 .
  • Variant activity towards a modified tetrapeptide, N-succinyl-Ala-Ala-Pro-Leu-p-nitroanilide (AAPL-p-NA) which undergoes a colorimetric change upon protease-mediated hydrolysis was also measured.
  • the activity of the variants at three different pH values (7, 5.5 and 4.5) was measured.
  • the activity of variants following a five minute heat treatment at 65° C. was also measured.
  • the activities observed for each property measured are shown in FIG. 17 .
  • PLSR partial least squares regression
  • the PLSR-based sequence activity model was used to assign a regression coefficient to each varied amino acid.
  • the predicted activity for a proteinase K variant was calculated by summing the regression coefficients for amino acid substitutions that are present in that variant. In this case, terms to account for interactions between the varied amino acids were not included, although this can also be done. See, for example, Aita et al., 2002, Biopolymers 64: 95-105.
  • FIG. 18 shows a correlation between the predictions of the sequence-activity model and the measured ability of heat-treated proteinase K variants to hydrolyze AAPL-p-NA.
  • sequence-activity model was tested for its ability to predict the activity of variants that have not been measured, or to identify amino acid substitutions that contribute positively to a specific protein property and that can then be experimentally combined.
  • sequence activity model for heat-tolerant hydrolyzers of AAPL-p-NA, the regression coefficients from the model were tested, as shown in FIG. 19 .
  • testing AAPL-p-NA hydrolysis at lowered pH (5.5 or 4.5) might be considered an appropriate surrogate for the low pH tolerance that will be required by an enzyme that is producing lactic acid from polylactide.
  • testing AAPL-p-NA hydrolysis following heat treatment may measure the stability that will be required for an enzyme that must resist the thermal stresses of incorporation into a plastic. Thermostability was expressed in three ways: (i) as the absolute level of activity remaining following heat treatment, (ii) as the activity remaining relative to the activity prior to heat treatment, and (iii) as the product of these two values. Having obtained values for each of these proteinase properties, the correlation between the properties was examine, and the amino acid substitutions that would be selected by each screen were compared.
  • the optimization procedures of the present invention are illustrated for an antibody that binds and neutralizes Respiratory Syncytial Virus (RSV).
  • RSV Respiratory Syncytial Virus
  • the sequence of one such antibody is publicly available (Genbank accession # AAF21612).
  • a significant benefit of the computational antibody design system using the methods described in this invention is that only relatively small numbers of variants need to be synthesized and tested. This allows the use of functional tests that are more comprehensive than binding assays. Viral neutralization for example, is an important antibody function but the sequence and structural determinants are poorly understood.
  • Methods used to identify substitutions in the framework and CDR regions of the heavy chain of the AAF21612 antibody sequence are as follows.
  • the sequence of the heavy chain of the AAF21612 antibody was aligned using the kabat numbering system with germline human ig heavy chain sequences retrieved from the VBase database. A total of 49 sequences were aligned. This alignment may not limited to germline human sequences.
  • all sequences that are in the same structural class as AAF21612 as defined by Chothia and Lesk Chothia and Lesk (Chothia and Lesk, 1986, EMBO Journal 5, 823-826) can be used.
  • Rule 1 a Align the sequences using kabat numbering and select all substitutions found in any of the germline sequences. Classify the substitutions into two categories: (i) substitutions found in the framework region and (ii) substitutions found in the CDR.
  • Rule 1 b Reconstruct a phylogenetic tree using the Clustal W software based on the amino acid alignment in the framework region. For each substitution, calculate the evolutionary proximity of the closest germline in which that substitution occurs.
  • the evolutionary proximity EP is calculated as follows:
  • n d is the number of amino acid differences between two sequences
  • n is the total number of amino acids in the protein.
  • d is the Poisson-corrected p-distance between two sequences
  • ln(1 ⁇ p) is the natural logarithm of the p-distance.
  • Rule 2 b For each position, calculate the site heterogeneity, that is a measure of the number of different amino acids present at that position.
  • the site heterogeneity is calculated as the number of different amino acids seen at a position in the set of homologs (SH).
  • f( ) is a mathematical function.
  • the function was the parameter in the parentheses multiplied by 1, but the use of functions allows different weights to be applied in subsequent cycles.
  • Score CDR f′(SE) ⁇ f′(SN) ⁇ f′ (SM), where f′( ) is a mathematical function.
  • the function f′( ) was the parameter in the parentheses multiplied by 1, but the use of functions f′( ) allows different weights to be applied in subsequent cycles.
  • the relative number of framework versus CDR substitution can be modulated.
  • a maximum number of framework and/or CDR substitutions in a variant is set.
  • the methods used to identify substitutions in the framework and CDR regions of the heavy chain of the RSV-19 antibody sequence are as follows.
  • the sequence of the heavy chain of the RSV-19 antibody was aligned using the kabat numbering system with germline human ig heavy chain sequences retrieved from VBase database This alignment may not limited to germline human sequences.
  • all human antibody sequences that are in the same structural class as AAF21612 as defined by Chothia and Lesk Chothia and Lesk, 1986, EMBO Journal 5, 823-826) can be used. A total of 45 sequences were aligned.
  • the sequences were processed and substitutions scored according to a modified version of the scheme shown in FIG. 3 .
  • the modified process is shown in FIG. 23 .
  • Rule 1 a Align sequences using kabat numbering and select all substitutions found in any of the germline sequences. Classify the substitutions into two categories: (i) substitutions found in the framework region and (ii) substitutions found in the CDR. Select only these substitutions and consider them separately.
  • Rule 1 b Reconstruct a phylogenetic tree using the Clustal W software based on the amino acid alignment in the framework region. For each substitution, calculate the evolutionary proximity of the closest germline in which that substitution occurs. The evolutionary proximity (EP) is calculated, where EP is as defined in Section 6.2.
  • Rule 2 b For each position calculate the site heterogeneity, that is a measure of the number of different amino acids present at that position.
  • the site heterogeneity is calculated as the number of different amino acids seen at a position in the set of homologs (SH).
  • Score FW f ( EP ) ⁇ f ( SH ) ⁇ f ( SE ) ⁇ f ( SN ) ⁇ f ( SM ),
  • f( ) is a mathematical function.
  • the function was the parameter in the parentheses multiplied by 1, but the use of functions allows different weights to be applied in subsequent cycles.
  • f′( ) is a mathematical function.
  • the function was the parameter in the parentheses multiplied by 1, but the use of functions allows different weights to be applied in subsequent cycles.
  • the relative number of framework versus CDR substitution can be modulated.
  • a maximum number of framework and/or CDR substitutions in a variant can be set.
  • substitutions of human residues in framework regions are preferred.
  • substitutions in the CDR are designed to retain the activity while changing the amino acid in framework region more biased towards human sequences.
  • aspects of the present invention can be implemented as a computer program product that comprises a computer program mechanism embedded in a computer readable storage medium.
  • the computer program product could contain the program modules and/or data structures shown in FIG. 2 .
  • These program modules may be stored on a CD-ROM, magnetic disk storage product, digital video disk (DVD) or any other computer readable data or program storage product.
  • the software modules in the computer program product may also be distributed electronically, via the Internet or otherwise, by transmission of a computer data signal (in which the software modules are embedded) on a carrier wave.

Landscapes

  • Health & Medical Sciences (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Physics & Mathematics (AREA)
  • Chemical & Material Sciences (AREA)
  • Engineering & Computer Science (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Biophysics (AREA)
  • Spectroscopy & Molecular Physics (AREA)
  • General Health & Medical Sciences (AREA)
  • Medical Informatics (AREA)
  • Theoretical Computer Science (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Biotechnology (AREA)
  • Evolutionary Biology (AREA)
  • Proteomics, Peptides & Aminoacids (AREA)
  • Genetics & Genomics (AREA)
  • Molecular Biology (AREA)
  • Organic Chemistry (AREA)
  • Immunology (AREA)
  • Analytical Chemistry (AREA)
  • Crystallography & Structural Chemistry (AREA)
  • Data Mining & Analysis (AREA)
  • Virology (AREA)
  • Medicinal Chemistry (AREA)
  • Biochemistry (AREA)
  • Artificial Intelligence (AREA)
  • Public Health (AREA)
  • Software Systems (AREA)
  • Evolutionary Computation (AREA)
  • Epidemiology (AREA)
  • Databases & Information Systems (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Pulmonology (AREA)
  • Bioethics (AREA)
  • Peptides Or Proteins (AREA)
  • Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
  • Management, Administration, Business Operations System, And Electronic Commerce (AREA)
US10/566,954 2003-08-01 2004-07-30 Systems and Methods for Antibody Engineering Abandoned US20080050357A1 (en)

Priority Applications (3)

Application Number Priority Date Filing Date Title
US10/566,954 US20080050357A1 (en) 2003-08-01 2004-07-30 Systems and Methods for Antibody Engineering
US12/726,843 US8412461B2 (en) 2003-08-01 2010-03-18 Systems and methods for antibody engineering
US13/854,002 US20140032186A1 (en) 2003-08-01 2013-03-29 Systems and methods for antibody engineering

Applications Claiming Priority (5)

Application Number Priority Date Filing Date Title
US49181503P 2003-08-01 2003-08-01
US53635704P 2004-01-14 2004-01-14
US53686204P 2004-01-15 2004-01-15
US10/566,954 US20080050357A1 (en) 2003-08-01 2004-07-30 Systems and Methods for Antibody Engineering
PCT/US2004/024751 WO2005012877A2 (fr) 2003-08-01 2004-07-30 Systeme et procede de construction d'anticorps

Related Parent Applications (1)

Application Number Title Priority Date Filing Date
PCT/US2004/024751 A-371-Of-International WO2005012877A2 (fr) 2003-08-01 2004-07-30 Systeme et procede de construction d'anticorps

Related Child Applications (1)

Application Number Title Priority Date Filing Date
US12/726,843 Continuation US8412461B2 (en) 2003-08-01 2010-03-18 Systems and methods for antibody engineering

Publications (1)

Publication Number Publication Date
US20080050357A1 true US20080050357A1 (en) 2008-02-28

Family

ID=34119818

Family Applications (3)

Application Number Title Priority Date Filing Date
US10/566,954 Abandoned US20080050357A1 (en) 2003-08-01 2004-07-30 Systems and Methods for Antibody Engineering
US12/726,843 Expired - Lifetime US8412461B2 (en) 2003-08-01 2010-03-18 Systems and methods for antibody engineering
US13/854,002 Abandoned US20140032186A1 (en) 2003-08-01 2013-03-29 Systems and methods for antibody engineering

Family Applications After (2)

Application Number Title Priority Date Filing Date
US12/726,843 Expired - Lifetime US8412461B2 (en) 2003-08-01 2010-03-18 Systems and methods for antibody engineering
US13/854,002 Abandoned US20140032186A1 (en) 2003-08-01 2013-03-29 Systems and methods for antibody engineering

Country Status (3)

Country Link
US (3) US20080050357A1 (fr)
EP (2) EP2434420A3 (fr)
WO (2) WO2005012877A2 (fr)

Cited By (12)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060136184A1 (en) * 2003-08-01 2006-06-22 Dna Twopointo Inc. Systems and methods for antibody engineering
WO2010006059A1 (fr) * 2008-07-08 2010-01-14 Abbott Laboratories Protéines de liaison à la prostaglandine e2 et leurs utilisations
US20110166844A1 (en) * 2003-08-01 2011-07-07 Dna Twopointo, Inc Systems and Methods for Antibody Engineering
CN102199575A (zh) * 2011-03-16 2011-09-28 中国检验检疫科学研究院 玉米褪绿斑驳病毒单克隆抗体及其应用
US20180144370A1 (en) * 2016-11-22 2018-05-24 Conversant, Llc Signal based ecpm prediction and bidding
US10223499B2 (en) * 2012-03-07 2019-03-05 Genformatic, Llc Method and apparatus for identification of biomolecules
US10438581B2 (en) * 2013-07-31 2019-10-08 Google Llc Speech recognition using neural networks
JP2020530991A (ja) * 2017-07-31 2020-11-05 エフ．ホフマン−ラ ロシュ アーゲーＦ． Ｈｏｆｆｍａｎｎ−Ｌａ Ｒｏｃｈｅ Ａｋｔｉｅｎｇｅｓｅｌｌｓｃｈａｆｔ 三次元構造に基づくヒト化方法
US11216742B2 (en) 2019-03-04 2022-01-04 Iocurrents, Inc. Data compression and communication using machine learning
US11288582B2 (en) * 2019-03-29 2022-03-29 Rovi Guides, Inc. Systems and methods for providing media content recommendations
CN115331728A (zh) * 2022-08-12 2022-11-11 杭州力文所生物科技有限公司 一种稳定折叠的富含二硫键的多肽设计方法及其电子设备
US11861732B1 (en) * 2022-07-27 2024-01-02 Intuit Inc. Industry-profile service for fraud detection

Families Citing this family (89)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US7662925B2 (en) 2002-03-01 2010-02-16 Xencor, Inc. Optimized Fc variants and methods for their generation
US7317091B2 (en) 2002-03-01 2008-01-08 Xencor, Inc. Optimized Fc variants
US20040132101A1 (en) 2002-09-27 2004-07-08 Xencor Optimized Fc variants and methods for their generation
US8188231B2 (en) 2002-09-27 2012-05-29 Xencor, Inc. Optimized FC variants
US8388955B2 (en) 2003-03-03 2013-03-05 Xencor, Inc. Fc variants
US8084582B2 (en) 2003-03-03 2011-12-27 Xencor, Inc. Optimized anti-CD20 monoclonal antibodies having Fc variants
US20090010920A1 (en) 2003-03-03 2009-01-08 Xencor, Inc. Fc Variants Having Decreased Affinity for FcyRIIb
US9051373B2 (en) 2003-05-02 2015-06-09 Xencor, Inc. Optimized Fc variants
US9714282B2 (en) 2003-09-26 2017-07-25 Xencor, Inc. Optimized Fc variants and methods for their generation
US8101720B2 (en) 2004-10-21 2012-01-24 Xencor, Inc. Immunoglobulin insertions, deletions and substitutions
CA2561264A1 (fr) 2004-03-24 2005-10-06 Xencor, Inc. Variantes d'immunoglobuline a l'exterieur de la region fc
US20150010550A1 (en) 2004-07-15 2015-01-08 Xencor, Inc. OPTIMIZED Fc VARIANTS
US8546543B2 (en) 2004-11-12 2013-10-01 Xencor, Inc. Fc variants that extend antibody half-life
EP2325207B1 (fr) 2004-11-12 2017-03-15 Xencor, Inc. Variants de fc avec une liaison altérée à fcrn
US8802820B2 (en) 2004-11-12 2014-08-12 Xencor, Inc. Fc variants with altered binding to FcRn
US8367805B2 (en) 2004-11-12 2013-02-05 Xencor, Inc. Fc variants with altered binding to FcRn
WO2007041635A2 (fr) 2005-10-03 2007-04-12 Xencor, Inc. Variants de fc dotés de propriétés de liaison aux récepteurs fc optimisées
AU2006302254B2 (en) 2005-10-06 2011-05-26 Xencor, Inc. Optimized anti-CD30 antibodies
CA2658557C (fr) 2006-08-14 2015-12-01 Xencor, Inc. Anticorps optimises ciblant cd19
AU2007299843B2 (en) 2006-09-18 2012-03-08 Xencor, Inc Optimized antibodies that target HM1.24
EP2118281A2 (fr) 2007-02-12 2009-11-18 Codexis, Inc. Relations structure-activite
DE102007011912A1 (de) * 2007-03-13 2008-09-18 Sanofi-Aventis Verfahren für das Erzeugen von Peptidbibliotheken und deren Verwendung
BR122021010656B1 (pt) 2007-12-26 2022-07-19 Xencor, Inc Anticorpo anti-tnf e composição farmacêutica
WO2009149218A2 (fr) * 2008-06-03 2009-12-10 Codon Devices, Inc. Nouvelles protéines et procédés de conception et d'utilisation de celles-ci
US7561972B1 (en) 2008-06-06 2009-07-14 Dna Twopointo, Inc. Synthetic nucleic acids for expression of encoded proteins
US7561973B1 (en) 2008-07-31 2009-07-14 Dna Twopointo, Inc. Methods for determining properties that affect an expression property value of polynucleotides in an expression system
US8401798B2 (en) 2008-06-06 2013-03-19 Dna Twopointo, Inc. Systems and methods for constructing frequency lookup tables for expression systems
US8126653B2 (en) 2008-07-31 2012-02-28 Dna Twopointo, Inc. Synthetic nucleic acids for expression of encoded proteins
US9493578B2 (en) 2009-09-02 2016-11-15 Xencor, Inc. Compositions and methods for simultaneous bivalent and monovalent co-engagement of antigens
WO2011091078A2 (fr) 2010-01-19 2011-07-28 Xencor, Inc. Variants d'anticorps possédant une activité complémentaire accrue
US20140297397A1 (en) * 2010-04-25 2014-10-02 Tomer Bakalash Use of neural activation in the superior temporal sulcus as a predictor for memorability of audio and/or visual content and emotional engagement
KR20120052683A (ko) * 2010-11-16 2012-05-24 한국전자통신연구원 지능형 서비스를 위한 다자간 상황정보 공유 장치 및 방법
US20140199318A1 (en) * 2011-08-12 2014-07-17 Max-Del-Bruck-Centrum Fur Molekulare Medizin Agonistic autoantibodies to the alpha1-adrenergic receptor and the beta2-adrenergic receptor in alzheimer's and vascular dementia
CN102870534B (zh) * 2012-09-21 2014-09-17 浙江大学 一种基于遗传算法的穴盘苗自动移钵路径优化方法
CN102854289B (zh) * 2012-10-11 2015-03-25 云南烟草科学研究院 一种初烤烟叶主流烟气中苯酚释放量的预测方法
US20140207267A1 (en) * 2013-01-23 2014-07-24 Hewlett-Packard Development Company, L.P. Metric based on estimate value
WO2015164723A1 (fr) 2014-04-25 2015-10-29 The Trustees Of The University Of Pennsylvania Procédés et compositions pour le traitement du cancer du sein métastatique et d'autres cancers dans le cerveau
DK3142750T3 (da) 2014-05-13 2020-09-14 Univ Pennsylvania Sammensætninger omfattende aav, som udtrykker dobbelt-antistofkonstrukter og anvendelser deraf
CN104262484A (zh) * 2014-10-17 2015-01-07 深圳出入境检验检疫局动植物检验检疫技术中心 一种抗非洲猪瘟病毒的特异性IgY抗体及制备方法和用途
US11535665B2 (en) 2015-05-13 2022-12-27 The Trustees Of The University Of Pennsylvania AAV-mediated expression of anti-influenza antibodies and methods of use thereof
CN108779167A (zh) 2015-10-28 2018-11-09 宾夕法尼亚州大学信托人 鞘内施用腺伴随病毒载体用于基因治疗
JP7347933B2 (ja) 2016-04-15 2023-09-20 ザ・トラステイーズ・オブ・ザ・ユニバーシテイ・オブ・ペンシルベニア 血友病ａ治療に対する遺伝子療法
US11191847B2 (en) 2016-04-15 2021-12-07 The Trustees Of The University Of Pennsylvania Gene therapy for treating hemophilia B
CN107688853B (zh) * 2016-08-05 2020-01-10 中科寒武纪科技股份有限公司 一种用于执行神经网络运算的装置及方法
KR102738752B1 (ko) 2016-12-30 2024-12-06 더 트러스티스 오브 더 유니버시티 오브 펜실베니아 페닐케톤뇨증을 치료하기 위한 유전자 치료
EP3562514A4 (fr) 2016-12-30 2020-08-05 The Trustees Of The University Of Pennsylvania Thérapie génique pour le traitement de la maladie de wilson
JP2020506705A (ja) 2017-02-01 2020-03-05 ザ・トラステイーズ・オブ・ザ・ユニバーシテイ・オブ・ペンシルベニア シトルリン血症を処置するための遺伝子療法
JOP20190200A1 (ar) 2017-02-28 2019-08-27 Univ Pennsylvania تركيبات نافعة في معالجة ضمور العضل النخاعي
SG10201912976WA (en) 2017-02-28 2020-02-27 Univ Pennsylvania Adeno-associated virus (aav) clade f vector and uses therefor
CN110573523A (zh) 2017-02-28 2019-12-13 宾夕法尼亚州大学信托人 基于aav载体的流感疫苗
RU2019130004A (ru) 2017-03-01 2021-04-01 Зе Трастис Оф Зе Юниверсити Оф Пеннсильвания Генная терапия при глазных заболеваниях
EP3634986A4 (fr) 2017-04-24 2021-09-08 The Trustees of The University of Pennsylvania Thérapie génique pour troubles oculaires
EP3621612A4 (fr) 2017-05-11 2021-06-16 The Trustees of The University of Pennsylvania Thérapie génique de céroïdes-lipofuscinoses neuronales
WO2018218359A1 (fr) 2017-05-31 2018-12-06 The Trustees Of The University Of Pennsylvania Thérapie génique destinée au traitement de troubles des peroxysomes
CN108168678B (zh) * 2017-11-22 2019-12-24 河南中烟工业有限责任公司 一种称重式条烟缺包检测动态数据处理方法
US11629198B2 (en) 2017-12-05 2023-04-18 The Trustees Of The University Of Pennsylvania Fusion proteins and antibodies targeting human red blood cell antigens
CN108102924B (zh) * 2018-01-12 2021-06-25 广东绿百多生物开发有限公司 一种利用噬菌蛭弧菌控制小球藻培养中细菌污染的方法
JP7498665B2 (ja) 2018-02-27 2024-06-12 ザ・トラステイーズ・オブ・ザ・ユニバーシテイ・オブ・ペンシルベニア 新規アデノ随伴ウイルス（ａａｖ）ベクター、低減されたカプシド脱アミド化を有するａａｖベクター、およびその使用
TW201939514A (zh) * 2018-03-16 2019-10-01 日商弘泰生物科技股份有限公司 免疫實體之有效率的分群
JP6500144B1 (ja) * 2018-03-28 2019-04-10 Ｋｏｔａｉバイオテクノロジーズ株式会社 免疫実体の効率的クラスタリング
CN109035229B (zh) * 2018-07-18 2022-02-18 河南牧业经济学院 一种基于傅里叶描述子的奶牛体况自动评价方法
US20210225455A1 (en) * 2018-08-15 2021-07-22 Zymergen Inc. Bioreachable prediction tool with biological sequence selection
US20230040603A1 (en) 2018-10-01 2023-02-09 The Trustees Of The University Of Pennsylvania Compositions useful for treating gm1 gangliosidosis
US11699094B2 (en) * 2018-10-31 2023-07-11 Salesforce, Inc. Automatic feature selection and model generation for linear models
JP7236253B2 (ja) * 2018-11-07 2023-03-09 合同会社Ｈ．Ｕ．グループ中央研究所 情報処理方法および学習モデル
WO2020102723A1 (fr) 2018-11-16 2020-05-22 Encoded Therapeutics, Inc. Compositions et méthodes pour le traitement de la maladie de wilson
DK3906066T5 (da) 2019-01-04 2024-08-05 Ultragenyx Pharmaceutical Inc Geneterapikonstruktioner til behandling af wilsons sygdom
SG11202108480UA (en) 2019-02-26 2021-09-29 Univ Pennsylvania Compositions useful in treatment of krabbe disease
CN111040031B (zh) * 2019-12-13 2020-11-03 中国农业科学院兰州兽医研究所 一种抗非洲猪瘟病毒的ScFv抗体及其制备方法
BR112022013914A2 (pt) 2020-02-02 2022-09-13 Univ Pennsylvania Composições úteis para tratar gangliosidose gm1
JP2023526311A (ja) 2020-05-12 2023-06-21 ザ・トラステイーズ・オブ・ザ・ユニバーシテイ・オブ・ペンシルベニア クラッベ病の治療に有用な組成物
CN111560069B (zh) * 2020-05-15 2021-07-02 中国农业科学院兰州兽医研究所 一种抗非洲猪瘟病毒的单链抗体及其制备方法和用途
TW202216760A (zh) 2020-07-23 2022-05-01 愛爾蘭商歐薩爾普羅席納有限公司 抗類澱粉β (ABETA)抗體
EP4196170A1 (fr) 2020-08-14 2023-06-21 The Trustees of The University of Pennsylvania Nouvelles capsides d'aav et compositions les contenant
KR20230083287A (ko) 2020-10-07 2023-06-09 리젠엑스바이오 인크. Cln2 질환의 안구 징후에 대한 유전자 요법
AU2021356684A1 (en) 2020-10-09 2023-05-11 The Trustees Of The University Of Pennsylvania Compositions and methods for treatment of fabry disease
WO2022094180A1 (fr) 2020-10-29 2022-05-05 The Trustees Of The University Of Pennsylvania Capsides de vaa et compositions les contenant
CA3209779A1 (fr) 2021-02-01 2022-08-04 Regenxbio Inc. Therapie genique de ceroides-lipofuscinoses neuronales
CN113185607B (zh) * 2021-03-22 2022-05-17 华南农业大学 一种多价禽腺病毒卵黄抗体的制备方法
WO2022232267A1 (fr) 2021-04-27 2022-11-03 The Trustees Of The University Of Pennsylvania Capsides du virus adéno-associé dérivé du porc et leurs utilisations
WO2023047557A1 (fr) 2021-09-27 2023-03-30 国立大学法人東北大学 Procédé de conception de protéine par apprentissage automatique
US20240384298A1 (en) 2021-10-02 2024-11-21 The Trustees Of The University Of Pennsylvania Novel aav capsids and compositions containing same
WO2023102517A1 (fr) 2021-12-02 2023-06-08 The Trustees Of The University Of Pennsylvania Compositions et méthodes de traitement de la maladie de fabry
US20250144243A1 (en) 2022-01-25 2025-05-08 The Trustees Of The University Of Pennsylvania Aav capsids for improved heart transduction and detargeting of liver
WO2023240236A1 (fr) 2022-06-10 2023-12-14 Voyager Therapeutics, Inc. Compositions et procédés pour le traitement de troubles liés à l'amyotrophie spinale
WO2024015966A2 (fr) 2022-07-15 2024-01-18 The Trustees Of The University Of Pennsylvania Vaa recombinants ayant des capsides de clade d et de clade e de vaa et compositions les contenant
WO2024105638A1 (fr) 2022-11-18 2024-05-23 Jcr Pharmaceuticals Co., Ltd. Vecteurs aav recombinants et méthodes de traitement de la maladie de hunter
JPWO2024122449A1 (fr) 2022-12-06 2024-06-13
WO2024241176A1 (fr) 2023-05-24 2024-11-28 Jcr Pharmaceuticals Co., Ltd. Vecteurs aav recombinés et procédés pour le traitement de maladies avec troubles du système nerveux central

Citations (8)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5843708A (en) * 1988-01-05 1998-12-01 Ciba-Geigy Corporation Chimeric antibodies
US20020048772A1 (en) * 2000-02-10 2002-04-25 Dahiyat Bassil I. Protein design automation for protein libraries
US20020119492A1 (en) * 2000-07-10 2002-08-29 Chirino Arthur J. Protein design automation for designing protein libraries with altered immunogenicity
US20020177170A1 (en) * 2001-04-17 2002-11-28 Peizhi Luo Structure-based selection and affinity maturation of antibody library
US20040072245A1 (en) * 2002-03-01 2004-04-15 Maxygen, Inc. Methods, systems, and software for identifying functional biomolecules
US20040161796A1 (en) * 2002-03-01 2004-08-19 Maxygen, Inc. Methods, systems, and software for identifying functional biomolecules
US20060136184A1 (en) * 2003-08-01 2006-06-22 Dna Twopointo Inc. Systems and methods for antibody engineering
US20070239364A1 (en) * 2002-03-01 2007-10-11 Maxygen, Inc. Methods, systems, and software for identifying functional biomolecules

Family Cites Families (5)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US4816567A (en) 1983-04-08 1989-03-28 Genentech, Inc. Recombinant immunoglobin preparations
GB2276169A (en) 1990-07-05 1994-09-21 Celltech Ltd Antibodies specific for carcinoembryonic antigen
US5681702A (en) 1994-08-30 1997-10-28 Chiron Corporation Reduction of nonspecific hybridization by using novel base-pairing schemes
WO2005012877A2 (fr) 2003-08-01 2005-02-10 Dna Twopointo Inc. Systeme et procede de construction d'anticorps
JP2006133755A (ja) * 2004-10-07 2006-05-25 Pentax Corp ズームレンズ系

Patent Citations (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US5843708A (en) * 1988-01-05 1998-12-01 Ciba-Geigy Corporation Chimeric antibodies
US20020048772A1 (en) * 2000-02-10 2002-04-25 Dahiyat Bassil I. Protein design automation for protein libraries
US20020119492A1 (en) * 2000-07-10 2002-08-29 Chirino Arthur J. Protein design automation for designing protein libraries with altered immunogenicity
US20020177170A1 (en) * 2001-04-17 2002-11-28 Peizhi Luo Structure-based selection and affinity maturation of antibody library
US7117096B2 (en) * 2001-04-17 2006-10-03 Abmaxis, Inc. Structure-based selection and affinity maturation of antibody library
US20040072245A1 (en) * 2002-03-01 2004-04-15 Maxygen, Inc. Methods, systems, and software for identifying functional biomolecules
US20040161796A1 (en) * 2002-03-01 2004-08-19 Maxygen, Inc. Methods, systems, and software for identifying functional biomolecules
US20060205003A1 (en) * 2002-03-01 2006-09-14 Maxygen, Inc. Methods, systems, and software for identifying functional biomolecules
US20070239364A1 (en) * 2002-03-01 2007-10-11 Maxygen, Inc. Methods, systems, and software for identifying functional biomolecules
US20060136184A1 (en) * 2003-08-01 2006-06-22 Dna Twopointo Inc. Systems and methods for antibody engineering

Cited By (25)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
US20060136184A1 (en) * 2003-08-01 2006-06-22 Dna Twopointo Inc. Systems and methods for antibody engineering
US20110059860A1 (en) * 2003-08-01 2011-03-10 Dna Twopointo Inc. Systems and methods for biopolymer engineering
US20110166844A1 (en) * 2003-08-01 2011-07-07 Dna Twopointo, Inc Systems and Methods for Antibody Engineering
US8005620B2 (en) 2003-08-01 2011-08-23 Dna Twopointo Inc. Systems and methods for biopolymer engineering
US8412461B2 (en) 2003-08-01 2013-04-02 Dna Twopointo, Inc. Systems and methods for antibody engineering
US8635029B2 (en) 2003-08-01 2014-01-21 Dna Twopointo, Inc. Systems and methods for biopolymer engineering
WO2010006059A1 (fr) * 2008-07-08 2010-01-14 Abbott Laboratories Protéines de liaison à la prostaglandine e2 et leurs utilisations
US20100040537A1 (en) * 2008-07-08 2010-02-18 Abbott Laboratories Prostaglandin E2 Binding Proteins and Uses Thereof
US8624002B2 (en) 2008-07-08 2014-01-07 Abbvie, Inc. Prostaglandin E2 binding proteins and uses thereof
CN104829718A (zh) * 2008-07-08 2015-08-12 艾伯维公司 前列腺素e2结合蛋白及其用途
CN102199575A (zh) * 2011-03-16 2011-09-28 中国检验检疫科学研究院 玉米褪绿斑驳病毒单克隆抗体及其应用
US10223499B2 (en) * 2012-03-07 2019-03-05 Genformatic, Llc Method and apparatus for identification of biomolecules
US11620991B2 (en) 2013-07-31 2023-04-04 Google Llc Speech recognition using neural networks
US10438581B2 (en) * 2013-07-31 2019-10-08 Google Llc Speech recognition using neural networks
US12243515B2 (en) 2013-07-31 2025-03-04 Google Llc Speech recognition using neural networks
US10930271B2 (en) 2013-07-31 2021-02-23 Google Llc Speech recognition using neural networks
US20180144370A1 (en) * 2016-11-22 2018-05-24 Conversant, Llc Signal based ecpm prediction and bidding
US11680110B2 (en) 2017-07-31 2023-06-20 Hoffmann-La Roche Inc. Three-dimensional structure-based humanization method
JP2020530991A (ja) * 2017-07-31 2020-11-05 エフ．ホフマン−ラ ロシュ アーゲーＦ． Ｈｏｆｆｍａｎｎ−Ｌａ Ｒｏｃｈｅ Ａｋｔｉｅｎｇｅｓｅｌｌｓｃｈａｆｔ 三次元構造に基づくヒト化方法
US11468355B2 (en) 2019-03-04 2022-10-11 Iocurrents, Inc. Data compression and communication using machine learning
US11216742B2 (en) 2019-03-04 2022-01-04 Iocurrents, Inc. Data compression and communication using machine learning
US11288582B2 (en) * 2019-03-29 2022-03-29 Rovi Guides, Inc. Systems and methods for providing media content recommendations
US12430569B2 (en) 2019-03-29 2025-09-30 Adeia Guides Inc. Systems and methods for providing media content recommendations
US11861732B1 (en) * 2022-07-27 2024-01-02 Intuit Inc. Industry-profile service for fraud detection
CN115331728A (zh) * 2022-08-12 2022-11-11 杭州力文所生物科技有限公司 一种稳定折叠的富含二硫键的多肽设计方法及其电子设备

Also Published As

Publication number Publication date
WO2005013090A2 (fr) 2005-02-10
WO2005012877A3 (fr) 2006-04-06
EP2434420A3 (fr) 2012-07-25
EP2434420A2 (fr) 2012-03-28
US20110166844A1 (en) 2011-07-07
US20140032186A1 (en) 2014-01-30
WO2005013090A3 (fr) 2006-04-13
WO2005012877A2 (fr) 2005-02-10
EP1660970A4 (fr) 2007-02-14
EP1660970A2 (fr) 2006-05-31
US8412461B2 (en) 2013-04-02

Similar Documents

Publication Publication Date Title
US8412461B2 (en) Systems and methods for antibody engineering
US8635029B2 (en) Systems and methods for biopolymer engineering
Cohen et al. NanoNet: Rapid and accurate end-to-end nanobody modeling by deep learning
US20220415436A1 (en) Computer assisted antibody re-epitoping
Willis et al. Human germline antibody gene segments encode polyspecific antibodies
Baran et al. Principles for computational design of binding antibodies
Obrezanova et al. Aggregation risk prediction for antibodies and its application to biotherapeutic development
Sircar et al. SnugDock: paratope structural optimization during antibody-antigen docking compensates for errors in antibody homology models
AU2004261198A1 (en) Altered antibodies having improved antigen-binding affinity
Bozhanova et al. Discovery of Marburg virus neutralizing antibodies from virus-naïve human antibody repertoires using large-scale structural predictions
Vishwakarma et al. VHH structural modelling approaches: a critical review
Petersen et al. Regulatory approved monoclonal antibodies contain framework mutations predicted from human antibody repertoires
Sevy et al. Antibodies: Computer‐aided prediction of structure and design of function
Lubin et al. Structural models of SARS-CoV-2 Omicron variant in complex with ACE2 receptor or antibodies suggest altered binding interfaces
Eshak et al. Advancements in nanobody epitope prediction: A comparative study of alphafold2multimer vs alphafold3
Licari et al. Embedding dynamics in intrinsic physicochemical profiles of market-stage antibody-based biotherapeutics
Sefid et al. In silico engineering towards enhancement of bap–VHH monoclonal antibody binding affinity
EP2795499A2 (fr) Maturation d'affinité in silico
Lee et al. Pairwise decomposition of residue interaction energies of single chain Fv with HIV-1 p17 epitope variants
Ketaren et al. Unique mechanisms to increase structural stability and enhance antigen binding in nanobodies
EP1939779A2 (fr) Systèmes et procédés d'ingénierie de biopolymère
Ferraz et al. Design of nanobody targeting SARS-CoV-2 spike glycoprotein using CDR-grafting assisted by molecular simulation and machine learning
del Alamo et al. Adapting ProteinMPNN for antibody design without retraining
Ketaren et al. Unique binding and stabilization mechanisms employed by and engineered into nanobodies
He et al. De novo generation of antibody CDRH3 with a pre-trained generative large language model

Legal Events

Date Code Title Description
AS Assignment

Owner name: DNA TWOPOINTO, INC., CALIFORNIA

Free format text: ASSIGNMENT OF ASSIGNORS INTEREST;ASSIGNORS:GUSTAFSSON, CLAES;GOVINDARAJAN, SRIDHAR;MINSHULL, JEREMY STEPHEN;REEL/FRAME:017907/0397;SIGNING DATES FROM 20060127 TO 20060130

STCB Information on status: application discontinuation

Free format text: ABANDONED -- FAILURE TO RESPOND TO AN OFFICE ACTION