CN116403639A - New antigen sequence generation method and system based on deep learning - Google Patents

Abstract

The invention discloses a new antigen sequence generation method and system based on deep learning, wherein the method comprises the following steps: acquiring original new antigen data; preprocessing original new antigen data to obtain an original new antigen sequence and a corresponding HLA sequence thereof; according to the original new antigen sequence and the corresponding HLA sequence, carrying out mutation on the original new antigen sequence, and carrying out Monte Carlo search on the original new antigen sequence in a sequence space to determine a mutated new antigen sequence which meets the design target; and obtaining a final new antigen sequence through screening of the new antigen sequence after mutation by molecular docking, presentation capability prediction and immunity capability prediction. The invention generates new antigen based on deep learning method, and screens the generated new antigen sequence, so that the finally obtained new antigen has better affinity and immunity.

Description

Method and system for generating neoantigen sequences based on deep learning

technical field

The present invention relates to the technical field of neoantigen sequence generation, in particular to a method and system for generating neoantigen sequences based on deep learning.

Background technique

Neoantigen, also known as neoantigen (Neoantigen), refers to a new protein formed on tumor cells (including malignant tumor cells, ie cancer cells) when certain mutations occur in the DNA of tumor cells. Neoantigens are selective in that they can elicit a T-cell response against tumors and thereby eliminate them, making neoantigens a key element in the design of cancer vaccines. Neoantigens play an important role in helping the body mount an immune response against cancer cells, and neoantigens used in vaccines and other types of immunotherapy are being studied to treat many types of cancer.

Existing technical solutions usually identify and identify immunogenic tumor neoantigens, which are algorithm designs based on existing neoantigens. At present, these neoantigens only occupy a small part of the protein space, and have not yet identified unknown tumor antigens. The vast space of proteins to explore. That is, the current research on neoantigens focuses on neoantigen recognition and neoantigen vaccine design, and there is a lack of related research on the generation of neoantigen sequences.

Contents of the invention

In order to solve the above-mentioned deficiencies in the prior art, the present invention provides a method and system for generating a neoantigen sequence based on deep learning. The method based on deep learning generates a brand new neoantigen, so that the generated neoantigen is comparable to the original neoantigen. Compared with human leucocyte antigen (HLA, human leucocyte antigen), it has a better affinity; by screening the generated neoantigen sequence, a neoantigen that can also be recognized and remembered by T cells can be obtained, and with subsequent biological verification, it can achieve Compared with existing neoantigens, better immunity can provide guidance value for the subsequent development of personalized vaccines.

In the first aspect, the present disclosure provides a method for generating new antigen sequences based on deep learning.

A method for generating neoantigen sequences based on deep learning, comprising:

Obtain raw neoantigen data;

Preprocess the original neoantigen data to obtain the original neoantigen sequence and its corresponding HLA sequence;

According to the original neoantigen sequence and its corresponding HLA sequence, the original neoantigen sequence is mutated, and a Monte Carlo search is performed on the original neoantigen sequence in the sequence space to determine the mutated neoantigen sequence that meets the design goal;

Screen the mutated neoantigen sequence through molecular docking, presentation ability prediction and immune ability prediction to obtain the final neoantigen sequence.

Further technical scheme, described pretreatment comprises:

Screen the original neoantigen data to screen out human neoantigen data, including neoantigen data and potential neoantigen data;

Filter potential neoantigen data to screen out potential neoantigen data with high confidence scores;

For each screened human neoantigen data, HLA-I type data is screened out, and the original neoantigen sequence and its corresponding HLA sequence of each data are obtained.

A further technical solution, said mutating the original neoantigen sequence according to the original neoantigen sequence and its corresponding HLA sequence, includes:

According to the original neoantigen sequence and its corresponding HLA sequence, the attention weight of each amino acid in the original neoantigen sequence to the HLA sequence is calculated, sorted according to the attention weight, and the amino acid position is selected for mutation.

In a further technical solution, the Monte Carlo search is performed on the original neoantigen sequence in the sequence space to determine the mutated neoantigen sequence that meets the design goal, including:

Input the mutated neoantigen sequence into the protein structure prediction model, and output the predicted protein structure;

Input the mutated neoantigen sequence and its corresponding HLA sequence into the polypeptide-HLA binding affinity prediction model, and output the predicted affinity;

Using the structural confidence and affinity score as the loss function, a Monte Carlo search is performed on the original neoantigen sequence in the sequence space to determine the final protein structure, and then the protein sequence design model is used to output the corresponding amino acid sequence.

In a further technical solution, the screening of the mutated neoantigen sequence through molecular docking, presentation ability prediction, and immune ability prediction to obtain the final neoantigen sequence includes:

The binding affinity between the generated neoantigen sequence and its corresponding HLA sequence is judged by molecular docking, and screening is performed according to the binding affinity to obtain a neoantigen sequence with higher affinity.

Further technical solutions also include:

Using the deep learning-based presentation ability model to capture the binding information of HLA-binding peptides, predict whether HLA-binding peptides can be displayed on the cell surface, and screen and predict neoantigen sequences that HLA-binding peptides can be displayed on the cell surface.

Further technical solutions also include:

Using the neoantigen characteristics as filtering criteria, the neoantigen characteristics are included in the immunogenicity score to calculate the immunogenicity score; the neoantigen characteristics include the dissociation constant and binding stability of the neoantigen-HLA complex and the mutated genes in the tumor expression;

The neoantigen with the highest score is used as the output result to obtain the final neoantigen sequence.

In a second aspect, the present disclosure provides a system for generating neoantigen sequences based on deep learning.

A system for generating neoantigen sequences based on deep learning, including:

A data acquisition module, configured to acquire raw neoantigen data;

The data preprocessing module is used to preprocess the original neoantigen data to obtain the original neoantigen sequence and its corresponding HLA sequence;

The neoantigen sequence design module is used to mutate the original neoantigen sequence based on the original neoantigen sequence and its corresponding HLA sequence, and perform a Monte Carlo search on the original neoantigen sequence in the sequence space to determine the mutation that meets the design goal After the neoantigen sequence;

The neoantigen sequence screening module is used to screen the mutated neoantigen sequence through molecular docking, presentation ability prediction and immune ability prediction to obtain the final neoantigen sequence.

In a third aspect, the present disclosure also provides an electronic device, including a memory, a processor, and computer instructions stored in the memory and run on the processor. When the computer instructions are executed by the processor, the computer instructions described in the first aspect can be completed. method steps.

In a fourth aspect, the present disclosure also provides a computer-readable storage medium for storing computer instructions, and when the computer instructions are executed by a processor, the steps of the method described in the first aspect are completed.

The above one or more technical solutions have the following beneficial effects:

1. The present invention provides a method and system for generating a neoantigen sequence based on deep learning. A new neoantigen is generated based on a deep learning method, so that the generated neoantigen is comparable to the human leukocyte antigen HLA compared with the original neoantigen. It has better affinity; by screening the generated neoantigen sequence, a neoantigen that can also be recognized and remembered by T cells is obtained, that is, a neoantigen with immune ability is obtained.

2. The present invention uses deep learning to design a brand new neoantigen, which greatly increases the exploration of protein space compared to the traditional scheme that only targets existing neoantigens.

3. The present invention cooperates with subsequent biological verification to achieve better immunity compared with existing neoantigens, and provides guidance for the subsequent development of personalized vaccines.

Description of drawings

The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention, and the schematic embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute improper limitations to the present invention.

FIG. 1 is an overall flow chart of the method for generating a new antigen sequence based on deep learning according to Embodiment 1 of the present invention;

Fig. 2 is a flow chart of the neoantigen design algorithm in Example 1 of the present invention.

Detailed ways

It should be noted that the following detailed description is exemplary and intended to provide further explanation of the present invention. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

It should be noted that the terminology used here is only for describing specific embodiments, and is not intended to limit exemplary embodiments according to the present invention. As used herein, unless the context clearly dictates otherwise, the singular is intended to include the plural, and it should also be understood that when the terms "comprising" and/or "comprising" are used in this specification, they mean There are features, steps, operations, means, components and/or combinations thereof.

Embodiment one

This embodiment provides a method for generating neoantigen sequences based on deep learning. First, the collected data in the original neoantigen database is preprocessed, and then the input neoantigen sequences are mutated by deep learning methods to obtain the corresponding HLA-related For better affinity and reasonable mutated neoantigen sequences, a series of methods are used to screen the generated neoantigens multiple times and cooperate with biological verification to obtain neoantigens with stronger immunity as the final result. The method described in this embodiment is shown in Figure 1, comprising the following steps:

Step S1, obtaining original neoantigen data;

Step S2, preprocessing the original neoantigen data to obtain the original neoantigen sequence and its corresponding HLA sequence;

Step S3. According to the original neoantigen sequence and its corresponding HLA sequence, mutate the original neoantigen sequence, and perform a Monte Carlo search on the original neoantigen sequence in the sequence space to determine the mutated neoantigen sequence that meets the design goal ;

Step S4, screening the mutated neoantigen sequence through molecular docking, presentation ability prediction and immune ability prediction to obtain the final neoantigen sequence.

First, in step S1, the original neoantigen data is obtained, and an original neoantigen data set is constructed. Given that accurate and complete datasets are highly correlated with the efficiency of deep learning, collecting high-quality datasets is an important start for generating neoantigen sequences. In this example, download data from NeoPeptide, TSNAdb, TransLnc and other databases, which respectively contain about 170,000 experimental neoantigens, 1.3 million and 400,000 potential neoantigens, and the distribution range includes melanoma, breast cancer, lung cancer, squamous cell carcinoma, etc.

Secondly, in step S2, the original neoantigen data is preprocessed to obtain the original neoantigen sequence and its corresponding HLA sequence. Before analyzing the input raw neoantigen data, these data need to be preprocessed. First of all, because the above database contains data of multiple species, such as humans, mice, etc., and this embodiment is for research on human neoantigens, therefore, the neoantigens of other species are filtered out to screen out human neoantigens, wherein the filtering The method includes: for each piece of data in the database, judging whether the HLA in each piece of data is human HLA (such as HLA-A, B, C, etc.), and sequentially judging whether the piece of data is from human beings. Then, because the data contains not only experimental neoantigens, but also potential neoantigens, for the potential neoantigen part, it is necessary to further filter to screen out potential neoantigens with high confidence scores, wherein the screening method includes: by using the neoantigen recognition tool ( Such as netMHCpan, etc.) to calculate the confidence score, the result obtained by using netMHCpan is %rank, if %rank<0.5, then it is considered that the new antigen has a higher affinity with the corresponding HLA, and the potential new antigen is screened out and selected as input. Finally, the HLA-I type data is screened out, and the required part is taken out of each piece of data, so that each piece of data only contains neoantigen sequences and their corresponding HLA sequences. The HLA sequence here refers to the amino acid sequence corresponding to HLA, and the following are expressed by HLA sequence.

Then, in step S3, the neoantigen sequence is designed, that is, the original neoantigen sequence is mutated according to the original neoantigen sequence and its corresponding HLA sequence, and a Monte Carlo search is performed on the original neoantigen sequence in the sequence space to determine that Design target mutated neoantigen sequences. The design algorithm is shown in Figure 2. The method described in this example is based on deep learning and extensively explores the protein space of the neoantigen by mutating the input original neoantigen sequence. This method does not have any restrictions on the structure of the original sequence.

Step S3.1. According to the original neoantigen sequence and its corresponding HLA sequence, calculate the attention weight of each amino acid in the original neoantigen sequence to the HLA sequence, sort according to the attention weight, and select amino acid positions for mutation.

The attention calculation is performed on the original neoantigen sequence and the corresponding HLA sequence, and the formula is:

Among them, K and V represent neoantigen sequences, and Q represents the corresponding HLA sequence. In the neoantigen sequence and HLA sequence, each character represents an amino acid, so the above parameters actually represent amino acid sequences. After the softmax operation in the formula, each amino acid of the neoantigen sequence gets its respective attention weight.

Afterwards, the intermediate calculation results are taken, that is, for HLA, the attention weights of each amino acid element of the neoantigen sequence are averaged, and the obtained value is used as the importance score of the corresponding amino acid. The higher the score, the corresponding amino acid may be combined with HLA. play a more important role. Rank the scores from low to high and only mutate the top ranked amino acid positions.

Step S3.2, input the mutated neoantigen sequence into the protein structure prediction model, output the predicted protein structure, input the mutated neoantigen sequence and its corresponding HLA sequence into the polypeptide-HLA binding affinity prediction model, and output the predicted Using the structural confidence and affinity score as the loss function, Monte Carlo search is performed on the original neoantigen sequence in the sequence space to determine the final protein structure, and then the protein sequence design model is used to output the corresponding amino acid sequence. The above-mentioned Monte Carlo search refers to Markov chain Monte Carlo optimization; the above-mentioned structure confidence refers to the two values of the model output plddt and ptm, which represent the difference between the output structure predicted by the model and the real structure. The intrinsic distance of the predicted structure, in the absence of natural structure, judges the reliability of each residue in the predicted structure; ptm represents the template modeling score predicted by the model, which evaluates the structure at the overall level.

A Monte Carlo search in sequence space was performed on the original neoantigen sequence, combined with a protein structure prediction model and a peptide-HLA binding affinity prediction model. The input of the protein structure prediction model is the amino acid sequence and the output is the predicted protein structure. The final output is obtained by extracting features from the sequence, encoding and decoding the features into the 3D position of the protein structure; the input of the peptide-HLA binding affinity prediction model is The peptide and the corresponding HLA, the output is the binding affinity of the polypeptide and the corresponding HLA, the scale of the training data is increased by using the binding affinity data and the elution ligand data, and the peptide-HLA binding affinity prediction model uses the NNAlign_MA architecture, so that the model can handle both types of data. In this example, during the training process of the above-mentioned protein structure prediction model and polypeptide-HLA binding affinity prediction model, the training data sets of the protein structure prediction model are the protein structure data set PDB and the protein sequence data set Uniclust30, and the polypeptide-HLA binding affinity The training data sets of the prediction model are the binding affinity data set and the elution ligand data set.

The output of the above two models is used as the loss function of the Monte Carlo search, and the sequence is continuously searched by combining Monte Carlo and the above model until the obtained structure meets the design goal, and finally the sequence is redesigned using the protein sequence design model. Contrary to the protein structure prediction model, the input is the protein structure and the output is the corresponding amino acid sequence. Through the message passing neural network, the use of encoder-decoder architecture and flexible decoding order makes the model have higher performance. The training data set of the above-mentioned protein sequence design model is the protein structure data set PDB. The sequence obtained by searching needs to use the protein structure prediction model to predict the structure. For each predicted structure, it is input to the protein sequence design model to generate the final corresponding sequence, so as to reduce overfitting in the search process.

Although the neoantigen obtained through the above design method is expected to better bind to HLA, because only a small amount of HLA-binding peptides can actually be displayed on the cell surface, it is impossible to obtain a neoantigen that can better stimulate immunity based on affinity alone , in this regard, this embodiment also adopts the following method for subsequent screening.

That is, in step S4, the screening of the mutated neoantigen sequence is carried out through molecular docking, presentation ability prediction and immune ability prediction to obtain the final neoantigen sequence.

After obtaining the amino acid sequence of the neoantigen through the above step S3, further screening of the obtained neoantigen sequence is still required. In the subsequent screening process, if structural data is required, the structure prediction model can be used to obtain the corresponding structure for screening, and finally determine the required neoantigen sequence after screening.

Among them, molecular docking is widely used in the study of protein-ligand interactions and in the discovery and development of drugs. Typically, on the basis of targets with known structures, molecular docking is used to predict the binding conformation and binding free energy of small molecules and targets to determine how they bind. In this example, molecular docking was used to determine the binding affinity of the designed and generated neoantigen sequence to its corresponding HLA sequence, and screening was performed based on the binding affinity to obtain a neoantigen sequence with higher affinity.

Neoantigens not only need to be able to bind to HLA, but also need to be presented on the cell surface in order to be recognized by T cells and trigger an immune response. In this regard, in this example, a deep learning-based presentation ability model is used to predict whether HLA-binding peptides can be displayed on the cell surface. The model makes predictions by capturing the binding information and circulation patterns of HLA-binding peptides. The above presentation ability model is trained with the eluted ligand data set as the training data set, which contains HLA monoallelic and multiallelic. Through this deep learning model, new antigens that cannot be presented on the cell surface are filtered out to further obtain more convincing results.

In order to further screen out neoantigens with immunocompetence, in this embodiment, the characteristics of neoantigens are used as filtering criteria, including the dissociation constant and binding stability of neoantigen-HLA complexes and the expression of mutant genes in tumors. The feature is incorporated into the immunogenicity score to calculate the score; the immunogenicity score is calculated for the neoantigen sequence, and the neoantigen with the highest score is used as the output result to obtain the final neoantigen sequence.

Through the above scheme of this embodiment, using deep learning to design a new neoantigen, compared with the traditional scheme that only targets existing neoantigens, greatly increases the exploration of protein space; and after using deep learning to design new antigens , and further screened the design results to obtain neoantigens with immunity, which made it more practical and could provide guidance for the subsequent development of personalized vaccines.

Embodiment two

This embodiment provides a system for generating neoantigen sequences based on deep learning, including:

A data acquisition module, configured to acquire raw neoantigen data;

The data preprocessing module is used to preprocess the original neoantigen data to obtain the original neoantigen sequence and its corresponding HLA sequence;

The neoantigen sequence design module is used to mutate the original neoantigen sequence based on the original neoantigen sequence and its corresponding HLA sequence, and perform a Monte Carlo search on the original neoantigen sequence in the sequence space to determine the mutation that meets the design goal After the neoantigen sequence;

The neoantigen sequence screening module is used to screen the mutated neoantigen sequence through molecular docking, presentation ability prediction and immune ability prediction to obtain the final neoantigen sequence.

Embodiment Three

This embodiment provides an electronic device, including a memory, a processor, and computer instructions stored in the memory and run on the processor. When the computer instructions are run by the processor, the above-mentioned new deep learning-based Steps in a method for generating an antigen sequence.

Embodiment four

This embodiment also provides a computer-readable storage medium for storing computer instructions, and when the computer instructions are executed by a processor, the steps in the method for generating neoantigen sequences based on deep learning as described above are completed.

The steps involved in the above embodiments 2 to 4 correspond to the method embodiment 1, and for specific implementation methods, please refer to the relevant description part of the embodiment 1. The term "computer-readable storage medium" shall be construed to include a single medium or multiple media including one or more sets of instructions; and shall also be construed to include any medium capable of storing, encoding, or carrying A set of instructions to execute and cause the processor to execute any method in the present invention.

Those skilled in the art should understand that each module or each step of the present invention described above can be realized by a general-purpose computer device, optionally, they can be realized by a program code executable by the computing device, thereby, they can be stored in a memory The device is executed by a computing device, or they are made into individual integrated circuit modules, or multiple modules or steps among them are made into a single integrated circuit module for realization. The invention is not limited to any specific combination of hardware and software.

The above descriptions are only preferred embodiments of the present invention, and are not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and changes. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present invention shall be included within the protection scope of the present invention.

Although the specific implementation of the present invention has been described above in conjunction with the accompanying drawings, it is not a limitation to the protection scope of the present invention. Those skilled in the art should understand that on the basis of the technical solution of the present invention, those skilled in the art do not need to pay creative work Various modifications or variations that can be made are still within the protection scope of the present invention.

Claims (10)

1. A method for generating neoantigen sequences based on deep learning, characterized in that it comprises: Obtain raw neoantigen data; Preprocess the original neoantigen data to obtain the original neoantigen sequence and its corresponding HLA sequence; According to the original neoantigen sequence and its corresponding HLA sequence, the original neoantigen sequence is mutated, and a Monte Carlo search is performed on the original neoantigen sequence in the sequence space to determine the mutated neoantigen sequence that meets the design goal; Screen the mutated neoantigen sequence through molecular docking, presentation ability prediction and immune ability prediction to obtain the final neoantigen sequence. 2. the new antigen sequence generation method based on deep learning as claimed in claim 1, is characterized in that, described preprocessing comprises: Screen the original neoantigen data to screen out human neoantigen data, including neoantigen data and potential neoantigen data; Filter potential neoantigen data to screen out potential neoantigen data with high confidence scores; For each screened human neoantigen data, HLA-I type data is screened out, and the original neoantigen sequence and its corresponding HLA sequence of each data are obtained. 3. The method for generating neoantigen sequences based on deep learning according to claim 1, wherein said mutating the original neoantigen sequences according to the original neoantigen sequences and their corresponding HLA sequences comprises: According to the original neoantigen sequence and its corresponding HLA sequence, the attention weight of each amino acid in the original neoantigen sequence to the HLA sequence is calculated, sorted according to the attention weight, and the amino acid position is selected for mutation. 4. The method for generating neoantigen sequences based on deep learning according to claim 1, wherein the original neoantigen sequence is subjected to a Monte Carlo search in sequence space to determine a mutated neoantigen that meets the design goal sequence, including: Input the mutated neoantigen sequence into the protein structure prediction model, and output the predicted protein structure; Input the mutated neoantigen sequence and its corresponding HLA sequence into the polypeptide-HLA binding affinity prediction model, and output the predicted affinity; Using the structural confidence and affinity score as the loss function, a Monte Carlo search is performed on the original neoantigen sequence in the sequence space to determine the final protein structure, and then the protein sequence design model is used to output the corresponding amino acid sequence. 5. The method for generating neoantigen sequences based on deep learning according to claim 1, wherein the screening of the mutated neoantigen sequences by molecular docking, presentation ability prediction, and immune ability prediction to obtain the final Neoantigen sequences, including: The binding affinity between the generated neoantigen sequence and its corresponding HLA sequence is judged by molecular docking, and screening is performed according to the binding affinity to obtain a neoantigen sequence with higher affinity. 6. The new antigen sequence generation method based on deep learning as claimed in claim 5, is characterized in that, also comprises: Using the deep learning-based presentation ability model to capture the binding information of HLA-binding peptides, predict whether HLA-binding peptides can be displayed on the cell surface, and screen and predict neoantigen sequences that HLA-binding peptides can be displayed on the cell surface. 7. The new antigen sequence generation method based on deep learning as claimed in claim 5, is characterized in that, also comprises: Using the neoantigen characteristics as filtering criteria, the neoantigen characteristics are included in the immunogenicity score to calculate the immunogenicity score; the neoantigen characteristics include the dissociation constant and binding stability of the neoantigen-HLA complex and the mutated genes in the tumor expression; The neoantigen with the highest score is used as the output result to obtain the final neoantigen sequence. 8. A neoantigen sequence generation system based on deep learning, characterized in that it comprises: A data acquisition module, configured to acquire raw neoantigen data; The data preprocessing module is used to preprocess the original neoantigen data to obtain the original neoantigen sequence and its corresponding HLA sequence; The neoantigen sequence design module is used to mutate the original neoantigen sequence based on the original neoantigen sequence and its corresponding HLA sequence, and perform a Monte Carlo search on the original neoantigen sequence in the sequence space to determine the mutation that meets the design goal After the neoantigen sequence; The neoantigen sequence screening module is used to screen the mutated neoantigen sequence through molecular docking, presentation ability prediction and immune ability prediction to obtain the final neoantigen sequence. 9. An electronic device, characterized in that it comprises a memory, a processor, and computer instructions stored in the memory and run on the processor, and when the computer instructions are run by the processor, any of claims 1-7 can be accomplished. The steps of a deep learning-based neoantigen sequence generation method described in one item. 10. A computer-readable storage medium, characterized in that it is used to store computer instructions, and when the computer instructions are executed by a processor, a deep learning-based Steps of a method for generating a neoantigen sequence.

Images