KR20130134901A - Method for predicting biological activity of unknown natural compounds using non-linear quantitative pattern-activity relationships model - Google Patents

Abstract

The present invention relates to a method for preparing a nonlinear biological activity prediction model of a natural unknown sample and a method for predicting the biological activity of a natural unknown sample. According to the biological activity prediction model of the natural product unknown sample and the method for predicting the biological activity of the natural product unknown sample according to an embodiment, it is possible to effectively evaluate the biological activity of the natural product unknown sample.

Description

Method for predicting biological activity of unknown natural compounds using non-linear quantitative pattern-activity relationships model}

The present invention relates to a method for preparing a nonlinear biological activity prediction model of a natural unknown sample and a method for predicting the biological activity of a natural unknown sample.

Drugs of herbal origin have long been used all over the world, especially in the three Asian countries of Japan, China and Japan. This is no different from now, in Korea, herbal medicines and extracts are used as therapeutic and prophylactic drugs in the form of herbal medicine, and herbal medicines and extracts are used as raw materials for natural medicines, health functional foods, and functional cosmetics processed through a certain process. It is widely used.

However, the quality control of the crude drugs that are the raw materials for these products, compared to such a wide range of uses, is very poor in its methods and standards. Currently, quality control of herbal medicines, herbal extracts, and herbal preparations is carried out in a manner to define the content range of one or two indicator components. However, the selection of the index component is a unique or large amount of ingredients present in the herbal medicine, which is different from the active ingredient in the herbal medicine, so this quality control standard is not appropriate as a quality control method for the herbal medicine as a pharmaceutical ingredient. Even if the index component is an active ingredient that directly affects the efficacy of the herbal medicine, the quality control method does not reflect the actual efficacy of the herbal medicine because of the properties of the herbal medicine that adds synergy and antagonism to the drug.

For this reason, many studies have been attempted to improve the quality control method of herbal medicines. Recently, the concept of phytoequivalence has been claimed in Germany that natural substances (or natural drugs and extracts) having certain physiological activities should be equivalent to those of the natural products to be compared. In addition, efficacy and safety can be indirectly assessed by evaluating chemical similarities between two natural products.In this case, chemical similarity should be achieved through pattern analysis of the entire chemical fingerprint rather than comparison with existing indicator components. Claims have been made and relevant research is being actively conducted.

Previous studies on Quantitative Pattern-Activity Relationships (QPARs) measured direct antioxidant capacity as a biological profile, and the predictive model through linear regression analysis using the PLS regression algorithm mainly focused on the antioxidant capacity of simple chemical binding. In contrast, the present invention performed the prediction of biological activity at the cellular level, and for this purpose, it can be meaningful that the first successful nonlinear regression and prediction using a random forest algorithm was performed.

The present invention provides a method for preparing a nonlinear biological activity prediction model of a natural unknown sample and a method for predicting the biological activity of a natural unknown sample.

One aspect includes obtaining a chemical finger print from a sample having known biological activity;

Measuring and quantifying the biological activity of the sample; And

It provides a method for producing a biological activity prediction model of an unknown sample comprising the step of obtaining the results of the chemical fingerprint and the numerical value of the biological activity using nonlinear regression and data mining techniques.

The present inventors have used a linear partial least square regression to correlate the known chemical fingerprints of natural products with their physiological activity data to predict simple physiological activities such as antioxidant capacity from natural chemical fingerprints. By introducing nonlinear data processing steps using optimal correlation shifting algorithms and random forest regression algorithms instead of the techniques reported in the previous work, we studied how to accurately predict the physiological activity of natural products even at more complex stages such as cell or animal experiments. The invention was completed.

The method for producing a biological activity prediction model of the unknown sample will be described in detail for each step as follows:

The method may first comprise obtaining a chemical fingerprint from a sample having known biological activity.

According to one embodiment, the sample may be selected from, for example, herbal, herbal, finished medicines and extracts or fractions thereof, but is not limited thereto.

In this step, it is preferable to prepare standard samples in order to set reference indicators, wherein the standard samples are of the same product having different chemical compositions (ie, herbal medicines and herbal medicines having the same plant or plant as the same plant of different origin). It is preferable to secure a certain number or more as an extract or the same drug product of different lots).

In addition, the extract is a variety of extraction solvents, for example, water, anhydrous or hydrous lower alcohol having 1 to 4 carbon atoms (methanol, ethanol, propanol, butanol, etc.), the mixed solvent of the lower alcohol and water, acetone, ethyl acetate , Chloroform or 1,3-butylene glycol can be obtained as an extraction solvent. In addition, the extract of the present invention may include not only the extract by the aforementioned extraction solvent, but also a fraction obtained by performing a conventional purification process or fractionating the extract according to the concentration of the solvent. For example, various purification methods additionally performed, such as separation using an ultrafiltration membrane having a constant molecular weight cut-off value, separation by various chromatography (manufactured for separation according to size, charge, hydrophobicity or affinity), etc. Fractions obtained through may also be included in the extract of the present invention.

On the other hand, the term "chemical fingerprint" is used herein to mean a specific chemical analysis pattern that can provide information about the chemical composition of the samples. According to one embodiment, the chemical fingerprint may be a chromatogram obtained from liquid chromatography, most preferably chromatogram obtained from high performance liquid chromatography (HPLC). Can be. In this case, the detector may be variously changed to UV or ELSD, MS, NMR, NIR, etc. according to the characteristics of the chemical component of the sample, and the analysis conditions are established to reflect the characteristics of the component in the sample to the maximum. It can later be determined as a criterion for chemical fingerprinting.

According to one embodiment, the method may further comprise the step of aligning the obtained chemical fingerprint after the obtaining step, for example, the correlation optimization (correlation optimized) of the obtained chemical fingerprint can be sorted using a warping algorithm.

In the case where the chemical fingerprinting is a chromatogram obtained from HPLC, in order to solve the error of the retention time occurring during the analysis of the chromatogram and to arrange the same time line, a peak alignment is performed. can do. According to one embodiment, the peak alignment may be performed using an optimal correlation shift algorithm for alignment of patterns without bias. Correlation Optimized Warping (COW) was developed by Nielson et al. In 1998 as an algorithm used to align the chemical fingerprint. In the analysis of the HPLC chromatogram, a slight change in the retention time may occur even in the same sample due to various factors, internal and external, and for the pattern analysis of the HPLC chromatogram, the change of the retention time is corrected to position the peaks between the samples. You need to align them equally.

According to one embodiment, the optimal correlation shift algorithm may have a segment length and a slack size as input parameters. The optimal correlation shift algorithm divides the sample chromatogram into several sections and then calculates a case where the similarity with the chromatogram to be aligned is maximized by shifting each section along the time axis. There are three input variables of the optimal correlation shift algorithm, which are the target chromatograms for the optimal correlation shift sorting, and the segment length, which is the length of each interval, and the slack size, which is the maximum warping distance. (slack size). For example, if the chromatogram to be sorted consists of 12000 data points, you can specify a segment length of 100 and a slack size of 30, where 12000 data points are 120 with a length of 100 data. Each section may be optimized as a case in which a correlation coefficient with a target chromatogram is maximized for cases in which 0 to 30 data points are shifted from side to side. After obtaining the optimized warping of each interval, the chromatogram alignment can be completed by filling the intermediate empty intervals. In general, (i) if the peaks are shifted over the original chromatogram and staggered, the slack size is reduced; If the peaks before the alignment are misaligned or misaligned after the alignment, the crack size can be increased while increasing the segment length. An example of Matlab code that can be used when performing this step is as follows.

<Matlab code>

# Import HPLC data. The raw data assumes the form of a txt or xlsx (excel) file.

for k = 1: n # (n: number of samples)

chr {p} = road ([ch- num2str (p) .txt]) For text files with the title of # ch1-.txt expression

    or

chr {p} = xlsread ([ch-num2str (p) .xlsx]) For an Excel file with the expression # chr-1.xlsx

end

timeline = chr {1} (1, :) # Store the timebase sequence of the chromatogram

chromatogram = zeros (n, ndata) (ndata: number of data points in the chromatogram)

for p = 1: n

chromatogram (p, :) = chr {p} (2, :) # Store the raw data of the chromatogram in a matrix.

end

cow = zeros (n, ndata)

for q = 1: n

[warping {p} cow (q, :) diagnos {p}] = cow (chromatogram (1, :) chromatogram (q, :) seg slack)

   # Perform COW on segment 1 and slacksize with target chromatogram 1

end

Next, the method may comprise measuring and quantifying the biological activity of the sample.

According to one embodiment, the measurement of biological activity may be to determine whether the survival of the animal, the prevention or treatment of a specific disease, whether tissue regeneration, whether the cell survival or the presence of enzyme activity, but is not limited thereto. Among the measurement methods, an experiment capable of sufficiently evaluating the activity of the herbal medicines, herbal extracts or finished drugs to be evaluated may be used as an index, wherein the evaluation target is any of the herbal medicines, herbal extracts or finished drugs. The concentration of the sample to be treated may vary. When evaluating a crude drug or extract, in principle, it is appropriate to treat extracts of the same concentration, and when performing lot-by-lot evaluation of a drug, based on the dose of the drug actually used (e.g. In the case of tablets, one tablet may be randomly selected from each lot) to determine the concentration of the sample to be treated. In addition, an assay may be performed using the prepared activity evaluation sample to quantify the activity of each sample. At this time, the obtained activity needs to sufficiently secure accuracy and precision statistically.

Finally, the method may include obtaining the chemical fingerprint and the quantified biological activity value using nonlinear regression and data mining techniques.

According to one embodiment, the nonlinear regression and data mining technique comprises a random forest regression algorithm, an artificial neural network algorithm, a support vector machine algorithm or a genetic algorithm. ), Or a modified form of the algorithm, but is not limited thereto.

In the aligning and quantifying step, values of aligned chemical fingerprints and quantified biological activities of samples to be used in the model design of the present invention can be obtained. Using this, a model can be designed using the Random Forest Regression Algorithm to evaluate the relationship between the biological activity and the data of chemical composition obtained from the chemical fingerprint of the sample.

The random forest regression algorithm is a machine learning algorithm developed by Breiman et al. In 2001. After a computer solves a problem, it remembers the reasoning process of problem solving and applies the experience to subsequent problem solving. Is one thing. The random forest regression algorithm consists of an ensemble of a large number of decision trees. For N samples having M variables, the random forest may form a model through the following algorithm.

(a) Bootstrapping Z * samples from N training data samples (random sampling allowing duplicates).

(b) The following process is repeated at the end node of the tree from the bootstrap data to form a tree.

i) Randomly extract m variables from M variables.

ii) Select the optimal variable / split-point from m variables.

iii) create two daughter nodes from this node

(c) Repeat this B times.

The result is predicted from the generated trees. When random forest is used in the regression analysis, the prediction result is derived as the arithmetic mean of the results of each tree.

 In this step, a model may be manufactured that uses the aligned chemical fingerprint as an input value and the quantified biological activity value as an output value, and the manufactured model may be represented as a matrix of various parameters.

According to one embodiment, the method may further comprise the step of determining the point of reference from the activity value output from the obtained result after the step of obtaining the result. The point of reference may vary depending on the biological activity measurement method used in the model.

Another aspect provides a computer readable recording medium for recording a computer program for executing a method of producing a biological activity prediction model of an unknown sample.

Embodiments according to the present invention may be implemented in the form of program instructions that may be executed by various computer components, and may be recorded in a computer-readable recording medium. The computer-readable recording medium may include program commands, data files, data structures, and the like, alone or in combination. Program instructions recorded on the computer-readable recording medium may be those specially designed and configured for the present invention, or may be known and available to those skilled in the computer software arts. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks and magnetic tape, optical recording media such as CD-ROMs, DVDs, and magneto-optical media such as floptical disks. media), and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, flash memory, and the like. Examples of program instructions include machine language code such as those generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter or the like. The hardware device may be configured to operate as one or more software modules for performing the processing according to the present invention, and vice versa.

Another aspect includes obtaining a chemical fingerprint from a sample having unknown biological activity; And

The obtained chemical fingerprint is applied to a biological activity prediction model of an unknown sample produced by the method, thereby obtaining a biological activity value of the sample.

The method for predicting the biological activity of the unknown sample in detail for each step is as follows:

The method may first comprise obtaining a chemical fingerprint from a sample having unknown biological activity.

According to one embodiment, the sample may be selected from the group consisting of herbal medicines, herbal medicines, finished medicines and extracts or fractions thereof, but the chemical fingerprints are chromatograms obtained from liquid chromatography, Preferably chromatogram obtained from high performance liquid chromatography. Extracts and chemical fingerprints are as described in the method for predicting the biological activity of the unknown sample.

Thereafter, the method may include applying the obtained chemical fingerprint to a biological activity prediction model of an unknown sample produced by the method to obtain a biological activity value of the sample.

According to one embodiment, the method may further comprise determining, after the obtaining step, a biological activity of a sample having an unknown biological activity from the obtained biological activity value.

Chemical fingerprints (eg HPLC chromatograms) of unknown samples obtained in the obtaining step may be substituted into biological activity prediction models of unknown samples prepared according to the above method to obtain biological activity values of unknown samples. The biological activity of the unknown sample can be assessed through whether the value satisfies the reference value.

According to the biological activity prediction model of the natural product unknown sample and the method for predicting the biological activity of the natural product unknown sample according to an embodiment, it is possible to effectively evaluate the biological activity of the natural product unknown sample.

FIG. 1 shows a series of chemical fingerprinting profiles using HPLC-DAD-MS for 40 seedling extracts.
Figures 2A and 2B show the results of sorting the HPLC-DAD data set of chemical fingerprinting for the leaf extract using the COW algorithm (Figure 2A shows the results before sorting and Figure 2B shows the results after sorting).
Figure 3 is a result of measuring the survival rate of aspirin-treated AGS cells to numerically represent the gastric protective function of the lobar samples.
Figure 4 is a graph showing the actual experimental values and prediction results of the biological activity of the leaf lobe using a random forest model.

Hereinafter, one or more embodiments will be described in more detail by way of examples. However, these embodiments are intended to illustrate one or more embodiments, and the scope of the present invention is not limited to these embodiments.

Example 1: Establishment of biological activity prediction model of unknown sample using leaflets and method for predicting biological activity of leaflet using same

The herbal extracts of 40 species of Artemisia princeps (35 domestic and 5 Chinese) were extracted as 95% ethanol extracts by using a pressurized solvent extraction method. Their chemical composition is shown as a series of chemical fingerprinting profiles using HPLC-DAD-MS (FIG. 1). At this time, the Correlation Optimized Warping (COW) algorithm was used to correct the retention time shift due to the characteristics of HPLC to match the standard profile, and the HPLC-DAD data set was aligned using the COW algorithm. The results are shown in Figure 2b. Then, in order to numerically represent the gastric protective function of each lobe sample, the survival rate of AGS cells toxic as aspirin was measured to obtain biological activity data of each sample (FIG. 3). The association of the data (chemical fingerprint data and biological activity data) was modeled using a random forest algorithm.

After dividing 40 types of leaf sample data into 30 training sets and 10 prediction sets, a random forest model was designed from the data of the training set, and then the HPLC fingerprint of the prediction set was extracted. The biological activity values (in this example, cell viability of AGS cells) were predicted by comparison with the actual experimental values, and the results are shown in FIG. 4.

As shown in FIG. 4, it was confirmed that a sufficient proportional relationship was established between the predicted value and the actual activity value. From these results, it was confirmed that the biological profile could be predicted from the chemical profile.

Claims (16)

Obtaining a chemical finger print from a sample having known biological activity;
Measuring and quantifying the biological activity of the sample; And
Obtaining the chemical fingerprint and the quantified biological activity value using nonlinear regression and data mining techniques to produce a biological activity prediction model of an unknown sample. The method of claim 1, wherein the method further comprises, after the obtaining step, aligning the obtained chemical fingerprint. The method of claim 2, wherein the aligning step is a stone for aligning the obtained chemical fingerprint using a correlation optimized warping algorithm. The method of claim 1, wherein the nonlinear regression and data mining technique comprises a random forest regression algorithm, an artificial neural network algorithm, a support vector machine algorithm, or a genetic algorithm. Method). The method of claim 1, wherein the sample is selected from the group consisting of herbal medicines, herbal medicines, finished medicines, and extracts or fractions thereof. The method of claim 1, wherein the chemical fingerprint is a chromatogram obtained from liquid chromatography. The method of claim 6, wherein the liquid chromatography is high performance liquid chromatography. 2. The method of claim 1, wherein the optimal correlation shift algorithm in the sorting step has segment length and slack size as input parameters. The method of claim 1, wherein the measurement of biological activity in the quantifying step measures whether the animal survives, prevents or treats a specific disease, whether tissue is regenerated, whether the cells survive, or the presence of enzyme activity. The method of claim 1, wherein the method further comprises determining, after obtaining the result, a point of reference among activity values output from the obtained result. A computer readable recording medium for recording a computer program for executing the method according to any one of claims 1 to 10. Obtaining a chemical fingerprint from a sample having unknown biological activity; And
A method for predicting the biological activity of an unknown sample, comprising applying the obtained chemical fingerprint to a model produced by the method of any one of claims 1 to 10 to obtain a biological activity value of the sample. The method of claim 12, wherein the sample is selected from the group consisting of herbal medicines, herbal medicines, finished medicines, and extracts or fractions thereof. The method of claim 12, wherein the chemical fingerprint is a chromatogram obtained from liquid chromatography. The method of claim 14, wherein the liquid chromatography is high performance liquid chromatography. The method of claim 12, wherein the method further comprises determining, after the obtaining step, a biological activity of a sample having an unknown biological activity from the obtained biological activity value.

Images