Interpretability analysis methods, devices, equipment, media and products for peptide activity

Abstract

The present invention relates to the field of polypeptide design technology, and discloses an interpretable analysis method, device, equipment, medium and product of polypeptide activity, the method comprising: obtaining polypeptide information to be analyzed corresponding to the polypeptide sample to be analyzed; based on a pre-established word element table, performing word element segmentation on each amino acid residue in the polypeptide sequence to obtain a polypeptide information word element set; inputting the polypeptide information word element set into a pre-established polypeptide activity prediction model to obtain a polypeptide activity category corresponding to the polypeptide sample to be analyzed; based on an interpretable sub-model of the polypeptide activity prediction model, using the polypeptide activity category as a calculation condition, determining the activity contribution value of the amino acid residues and modification information in the polypeptide sample to be analyzed. The present invention intuitively explains the degree of influence of the polypeptide sequence and modification information on the polypeptide activity, clarifies the specific contribution of each amino acid residue and its modification to the polypeptide activity, and is conducive to providing accurate polypeptide activity evaluation and improving the quality of polypeptide design.

Description

Technical Field

[0001] The present invention relates to the technical field of polypeptide design, and in particular to an interpretable analysis method, device, equipment, medium and product of polypeptide activity. Background Art

[0002] A peptide is a molecule formed by connecting multiple amino acids through peptide bonds. It has antiviral and antibacterial properties and plays an important role in drug development. When designing a peptide, it is often necessary to analyze the effect of the peptide sequence and modification on the activity of the peptide, which can help design active peptides and improve the quality of peptide design.

[0003] However, the inventors have found that most of the current methods for analyzing polypeptide sequences and the effects of modifications on polypeptide activity are unable to comprehensively analyze the effects of each amino acid residue in the polypeptide sequence and the degree of effects of amino acid residue modifications on polypeptide activity, thereby affecting the effectiveness of drug development. Summary of the invention

[0004] In view of this, the present invention provides a method, device, equipment, medium and product for interpretable analysis of polypeptide activity to comprehensively analyze the effects of polypeptide sequence and modification on polypeptide activity.

[0005] In a first aspect, the present invention provides a method for interpretable analysis of polypeptide activity, the method comprising:

[0006] Obtaining the polypeptide information to be analyzed corresponding to the polypeptide sample to be analyzed, the polypeptide information to be analyzed including polypeptide sequence and modification information;

[0007] Based on the pre-established word-unit table, each amino acid residue in the polypeptide sequence is segmented into words to obtain a polypeptide information word-unit set;

[0008] Inputting the polypeptide information word unit set into a pre-established polypeptide activity prediction model to obtain the polypeptide activity category corresponding to the polypeptide sample to be analyzed;

[0009] Based on the interpretable sub-model of the peptide activity prediction model, the peptide activity category is used as the calculation condition to determine the activity contribution value corresponding to the amino acid residues and modification information in the peptide sample to be analyzed.

[0010] In this embodiment, based on the trained activity prediction model, an interpretable method is established using the SHAP framework to calculate the activity contribution value of the polypeptide sequence and modification information corresponding to the sample to be analyzed, which intuitively explains the degree of influence of the polypeptide sequence and modification information on the polypeptide activity, and can clarify the specific contribution of each amino acid residue and its modification to the polypeptide activity, which is conducive to providing accurate polypeptide activity evaluation and improving the quality of polypeptide design.

[0011] In an optional embodiment, there are multiple activity contribution values ​​corresponding to the amino acid residues and the modification information. After determining the activity contribution values ​​corresponding to the amino acid residues and the modification information in the polypeptide sample to be analyzed, the following further comprises:

[0012] Determine the influencing factors to be analyzed, including: the overall impact of the peptide information to be analyzed on the peptide activity, the maximum impact of the peptide information to be analyzed on the peptide activity, and the positive and negative impact relationship between the peptide information to be analyzed and the peptide activity;

[0013] According to the influencing factors to be analyzed, the activity contribution values ​​corresponding to the amino acid residues and the modification information are calculated to determine the single activity contribution values ​​corresponding to the amino acid residues and the modification information;

[0014] The amino acid residues and modification information are ranked based on the single activity contribution value corresponding to the amino acid residues and modification information;

[0015] The sorting results and the corresponding activity contribution values ​​are visualized.

[0016] In this embodiment, after analyzing the activity contribution values ​​corresponding to the amino acid residues and modification information, the importance of different amino acid residues and modification information on the activity type of the polypeptide can be displayed by drawing a picture, which can not only show the overall influence and maximum influence of the polypeptide sequence and modification on the activity information, but also show the positive and negative relationship between the polypeptide sequence and modification and the activity information. In this way, the key factors contributing to the activity of the polypeptide can be clearly identified, which is helpful for optimizing the polypeptide design and functional research, and the sorting results and activity contribution values ​​can be visualized, making the analysis results more intuitive and easy to understand.

[0017] In an optional embodiment, the polypeptide activity prediction model is established by the following steps:

[0018] Acquire test polypeptide data corresponding to a plurality of test polypeptide samples, wherein the test polypeptide data includes test polypeptide sequence, test modification information, and test activity information;

[0019] Obtaining the position information corresponding to each amino acid residue in the test polypeptide sequence;

[0020] Based on the word-gram table, each amino acid residue in the test polypeptide sequence is segmented into words to obtain a test word-gram set;

[0021] Based on the position information and the test word set, an embedding vector corresponding to each test peptide sample is generated;

[0022] Input the embedding vector into the encoder structure and output the encoding result, wherein the encoder structure includes a multi-head self-attention layer, a first normalization and residual connection layer, a feedforward fully connected layer, and a second normalization and residual connection layer connected in sequence;

[0023] The encoding results are input into the hidden layer to extract the sequence tags corresponding to each test peptide sample;

[0024] The sequence tags are input into the classifier, and the model is trained based on the experimental activity information to generate a peptide activity prediction model.

[0025] In an optional embodiment, after generating the polypeptide activity prediction model, the following steps are included:

[0026] Load the peptide activity prediction model;

[0027] According to the model type of the peptide activity prediction model, an interpretability framework is created to establish an interpretable sub-model of the peptide activity prediction model.

[0028] In this embodiment, it is necessary to collect polypeptide sequences and modification information and corresponding activity values, then segment the sequences and modification information into tokens, convert them into digital signals, then build and train a polypeptide activity prediction model, input the digital signals into the model for training and testing, and save the optimal activity prediction model; then build an interpretability method for the prediction model, based on the trained activity prediction model, use the SHAP framework to establish an interpretability method, calculate the SHAP value of the polypeptide sequence and modification information, explain the relationship between the polypeptide sequence and modification information and the activity value, and can be based on all SHAP values ​​of the interpretation data set, and draw a picture to show the importance of different polypeptide sequence tokens and modification tokens to the polypeptide activity type. The present invention can comprehensively analyze the influence of polypeptide sequence and modification information on polypeptide activity, and can help design active polypeptides.

[0029] In an optional embodiment, based on the interpretability sub-model of the polypeptide activity prediction model, the polypeptide activity category is used as a calculation condition to determine the activity contribution value corresponding to the amino acid residues and modification information in the polypeptide sample to be analyzed, including:

[0030] When the peptide activity prediction model is a binary classification model, the active category is specified as the calculation condition;

[0031] When the polypeptide activity prediction model is a multi-classification model, the highest activity class or the lowest activity class is specified as the calculation condition.

[0032] In this embodiment, in the binary classification model, the effect of amino acid residues on the activity category can be directly evaluated by specifying the active category. In the multi-classification model, by specifying the highest or lowest activity category, the key influencing factors of a specific category can be understood more accurately, which is conducive to improving the effectiveness of peptide design.

[0033] In an optional implementation, the word table is established by the following steps:

[0034] Define the letter abbreviations corresponding to various amino acid residues one by one to obtain a basic word unit comparison table;

[0035] Define a one-to-one correspondence between multiple modification information and special modification words, and obtain a comparison table of special modification words;

[0036] Define special tokens, which include: completion token, start token, and end token;

[0037] Define unknown tokens;

[0038] The basic word unit comparison table, the modified special word unit comparison table, the marked special word unit, and the unknown word unit are mapped to numbers to establish a word unit table.

[0039] In this embodiment, the amino acid residues, modification information, and special words can be standardized into a unified representation through the word table, thereby improving the efficiency and consistency of model processing.

[0040] In a second aspect, the present invention provides an interpretable analysis device for polypeptide activity, the device comprising:

[0041] An acquisition module is used to obtain the polypeptide information to be analyzed corresponding to the polypeptide sample to be analyzed, and the polypeptide information to be analyzed includes polypeptide sequence and modification information;

[0042] A word unit segmentation module is used to perform word unit segmentation on each amino acid residue in the polypeptide sequence based on a pre-established word unit table, and obtain a polypeptide information word unit set in combination with modification information;

[0043] A prediction module is used to input the polypeptide information word set into a pre-established polypeptide activity prediction model to obtain the polypeptide activity category corresponding to the polypeptide sample to be analyzed;

[0044] The analysis module is used to determine the activity contribution values ​​of the amino acid residues and modification information in the peptide sample to be analyzed based on the interpretable sub-model of the peptide activity prediction model, taking the peptide activity category as the calculation condition.

[0045] In a third aspect, the present invention provides a computer device, comprising: a memory and a processor, the memory and the processor being communicatively connected to each other, the memory storing computer instructions, and the processor executing the computer instructions to perform the interpretable analysis method of polypeptide activity of the first aspect or any corresponding embodiment thereof.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium storing computer instructions for causing a computer to execute the interpretable analysis method of polypeptide activity of the first aspect or any corresponding embodiment thereof.

[0047] In a fifth aspect, the present invention provides a computer program product, comprising computer instructions for causing a computer to execute the method for interpretable analysis of polypeptide activity of the first aspect or any corresponding embodiment thereof.

[0048] It should be noted that the interpretable analysis device for polypeptide activity, the computer device and the computer-readable storage medium provided by the present invention correspond to the above-mentioned interpretable analysis method for polypeptide activity. Therefore, for the beneficial effects of the interpretable analysis device for polypeptide activity, the computer device and the computer-readable storage medium, please refer to the description of the corresponding beneficial effects of the interpretable analysis method for polypeptide activity above, which will not be repeated here. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the specific implementation methods of the present invention or the technical solutions in the prior art, the drawings required for use in the specific implementation methods or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are some implementation methods of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0050] Figure 1 is a schematic diagram of a process for interpretable analysis of polypeptide activity according to an embodiment of the present invention;

[0051] Figure 2 is a schematic diagram of the SHAP value results of a single polypeptide sequence according to an embodiment of the present invention;

[0052] Figure 3 is a visualization diagram of the overall effect of polypeptide sequence and modification on polypeptide activity according to an embodiment of the present invention;

[0053] Figure 4 is a visualization diagram of the maximum effect of polypeptide sequence and modification on polypeptide activity according to an embodiment of the present invention;

[0054] Figure 5It is a visualized schematic diagram of the positive and negative influence relationship between the polypeptide sequence and modification and the polypeptide activity according to an embodiment of the present invention;

[0055] Figure 6 is a schematic diagram of downloading an original data table from a DRAVP database according to an embodiment of the present invention;

[0056] Figure 7 is a schematic diagram of a process of model construction according to an embodiment of the present invention;

[0057] Figure 8 is a schematic diagram of a digital word table according to an embodiment of the present invention;

[0058] Fig. 9 is a structural block diagram of an interpretable analysis device for polypeptide activity according to an embodiment of the present invention;

[0059] Fig.10 It is a schematic diagram of the hardware structure of a computer device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0060] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0061] According to an embodiment of the present invention, an embodiment of a method for interpretable analysis of polypeptide activity is provided. It should be noted that the steps shown in the flowchart of the accompanying drawings can be executed in a computer system such as a set of computer executable instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in an order different from that shown here.

[0062] In this embodiment, a method for interpretable analysis of polypeptide activity is provided, which can be executed by a server, a terminal, a mobile terminal, etc. Figure 1 is a flow chart of a method for interpretable analysis of polypeptide activity according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0063] Step S101, obtaining the polypeptide information to be analyzed corresponding to the polypeptide sample to be analyzed, the polypeptide information to be analyzed including polypeptide sequence and modification information. The polypeptide sample to be analyzed may be a polypeptide extracted from a cell line, or may be a tissue sample or a synthetic sample, etc. Furthermore, mass spectrometry may be used to perform polypeptide sequence analysis and modification analysis, etc., or related polypeptide sequences and modifications may be queried from a known database.

[0064] Step S102, based on the pre-established word-gram table and the polypeptide information to be analyzed, word-gram segmentation is performed on each amino acid residue in the polypeptide sequence, and the polypeptide information word-gram set is obtained by combining the modification information.

[0065] In this embodiment, the established word table can be used directly, and the existing word table needs to include all possible words. After obtaining the polypeptide information to be analyzed, the polypeptide sequence and modification information can be segmented into words, wherein the polypeptide sequence can be segmented into words according to k-mer analysis, and the polypeptide sequence can also be directly segmented in the manner of each amino acid residue as a word. The modification information is not segmented, but directly used as a modified special word, thereby forming a polypeptide information word set corresponding to the polypeptide sample to be analyzed. The following example illustrates that the polypeptide sequence is: SRSELIVHQRRC, the polypeptide sequence SRSELIVHQRRC is segmented into words, and the special word is added. <cls>and <end>After that, the peptide information word set is obtained:' <cls>', 'S', 'R', 'S', 'E', 'L', 'I', 'V', 'H', 'Q', 'R', 'R', 'C', ' <end>'.

[0066] Step S103, inputting the polypeptide information word-unit set into a pre-established polypeptide activity prediction model to obtain the polypeptide activity category corresponding to the polypeptide sample to be analyzed.

[0067] In this embodiment, a polypeptide activity prediction model can be pre-built and trained, which can be a Transformer Encoder model trained from scratch. The model can be a binary classification model, that is, predicting whether the polypeptide is active. In addition, the model can also be a multi-classification model, that is, predicting multiple levels or multiple categories of polypeptide activity. After the polypeptide information word set is input into the polypeptide activity prediction model, it can be determined whether the corresponding polypeptide sample to be analyzed is in the active category or the inactive category.

[0068] Step S104, based on the interpretability sub-model of the polypeptide activity prediction model, the polypeptide activity category is used as a calculation condition to determine the activity contribution value of the amino acid residues and modification information in the polypeptide sample to be analyzed.

[0069] In this embodiment, the interpretability sub-model of the constructed polypeptide activity prediction model can be a custom prediction function built on the SHAP (SHapley Additive exPlanations, referred to as SHAP, SHAP is a method for explaining the prediction results of machine learning models) framework, and a category needs to be specified in the prediction function as a calculation condition; for a two-classification model, the positive category (i.e., the active category) can be specified as the calculation condition, that is, the SHAP value under the active condition can be calculated (the SHAP value indicates the degree of influence of a certain feature on the model prediction, that is, given a current set of feature values, the contribution of the feature value to the difference between the actual predicted value and the average predicted value); for a multi-classification model, the highest category (i.e., the highest activity level) can be specified as the calculation condition, that is, the SHAP value under the highest activity condition can be calculated; for a multi-classification model, the lowest category (i.e., the lowest activity level, inactive) can also be specified as the calculation condition, that is, the SHAP value under the inactive condition can be calculated. Among them, the SHAP value is the activity contribution value in this embodiment.

[0070] like Figure 2 The following is the SHAP value of the peptide sequence SRSELIVHQRRC. The model input is: <cls>', 'S', 'R', 'S', 'E', 'L', 'I', 'V', 'H', 'Q', 'R', 'R', 'C', ' <end>', that is Figure 2 After the input is entered into the model, the SHAP value under active conditions is calculated, which is Figure 2 According to the SHAP value corresponding to each amino acid residue and modification information, the influence of each amino acid residue and amino acid residue modification on the activity of the polypeptide can be comprehensively analyzed, which is of great significance to the research and development of drugs.

[0071] In this embodiment, based on the trained activity prediction model, an interpretable method is established using the SHAP framework to calculate the activity contribution value of the polypeptide sequence and modification information corresponding to the sample to be analyzed, and the degree of influence of the polypeptide sequence and modification information on the polypeptide activity is intuitively explained. The specific contribution of each amino acid residue and its modification to the polypeptide activity can be clarified, which is conducive to providing accurate polypeptide activity evaluation and improving the quality of polypeptide design.

[0072] In some optional embodiments, there are multiple activity contribution values ​​corresponding to the amino acid residues and the modification information. After determining the activity contribution values ​​corresponding to the amino acid residues and the modification information in the polypeptide sample to be analyzed, the following steps are further included:

[0073] Step S201, determining the influencing factors to be analyzed, the influencing factors to be analyzed include: the overall influence of the polypeptide information to be analyzed on the polypeptide activity, the maximum influence of the polypeptide information to be analyzed on the polypeptide activity, and the positive and negative influence relationship between the polypeptide information to be analyzed and the polypeptide activity.

[0074] Step S202, calculating the activity contribution values ​​corresponding to the amino acid residues and the modification information one by one according to the influencing factors to be analyzed, and determining a single activity contribution value corresponding to the amino acid residues and the modification information one by one.

[0075] Step S203, sorting the amino acid residues and the modification information based on the single activity contribution values ​​corresponding one to one between the amino acid residues and the modification information.

[0076] Step S204: Visually display the ranking results and the corresponding activity contribution values.

[0077] The activity contribution value corresponding to the analyzed amino acid residues and modification information is also called SHAP value, which is generally a SHAP value matrix. Each data instance in the SHAP value matrix has one row and each feature has one column. By combining the SHAP values ​​and determining a single activity contribution value, a global interpretation can be performed to explain the overall impact of the input polypeptide sequence and modification on the activity of the polypeptide. Figure 3 As shown, the visualization shows the overall effect of the input sequence and modification on the activity of the peptide. The amino acid residues and modifications that have the overall effect on the activity of the peptide are: C, G, S, K, D, P, T, I, A, F, V, R, L, Q, N, W, Y, H, modification 2, modification 1, E, M.

[0078] By taking the maximum SHAP value for each feature, the maximum effect of the input sequence and modification on the activity of the peptide can be explained. Figure 4 As shown in the figure, the input peptide sequence and the modification with the greatest impact on the peptide activity are visualized. The amino acid residues and modifications with the greatest impact on the peptide activity are: C, G, D, S, A, P, I, K, V, T, F, R, Q, L, W, H, N, Y, modification 1, modification 2, E, M.

[0079] By taking the average of the SHAP values ​​of each feature, we can get the positive and negative impact of the SHAP feature. Figure 5 As shown, the positive and negative effects of the input peptide sequence and modification on the peptide activity are visualized. The amino acid residues and modifications that have a positive effect on the peptide activity are: C, D, W, F, Q, N, Y, T, L, M. The amino acid residues and modifications that have a negative effect on the peptide activity are: G, modification 1, P, A, V, S, K, H, I, R, E, modification 2.

[0080] Among them, the explanation of the relationship between the polypeptide sequence and modification information and the polypeptide activity can be based on the SHAP value of a single input information or the SHAP value of multiple input information; when based on the SHAP value of a single polypeptide sequence and activity information, it explains the influence of different sequences and modifications in a single input on the activity; when based on the SHAP value of multiple polypeptide sequences and activity information, it explains the influence of different sequences and modifications in multiple inputs on the activity.

[0081] In this embodiment, after analyzing the activity contribution values ​​corresponding to the amino acid residues and modification information, the importance of different amino acid residues and modification information on the activity type of the polypeptide can be displayed by drawing a picture, which can not only show the overall influence and maximum influence of the polypeptide sequence and modification on the activity information, but also show the positive and negative relationship between the polypeptide sequence and modification and the activity information. In this way, the key factors contributing to the activity of the polypeptide can be clearly identified, which is helpful for optimizing the polypeptide design and functional research, and the sorting results and activity contribution values ​​can be visualized, making the analysis results more intuitive and easy to understand.

[0082] In some optional embodiments, the polypeptide activity prediction model is established by the following steps:

[0083] Step a1, obtaining test polypeptide data corresponding to a plurality of test polypeptide samples, wherein the test polypeptide data includes test polypeptide sequence, test modification information, and test activity information;

[0084] The data set used for training the model in this embodiment includes peptide sequence and modification and activity value information. In specific experiments, peptide data in the AVPdb database, peptide data in the DRAVP database, and peptide test data that have been tested by R&D personnel can be used to obtain a total of 7255 data as a data set. Figure 6 As shown in the figure, the original data table was downloaded from the DRAVP database. The original data table contains detailed information such as DRAVP_ID (peptide identification number), Sequence (peptide sequence and modification), Sequence_Length (sequence length), Name (name), Source (source), Uniprot_ID (Uniprot database identification number), PDB_ID (PDB database identification number), Target_Organism (target virus), Family (virus family), Assay (analysis method), Activity (activity), etc. Peptide sequence and activity information were extracted from the original table. The data set contains peptide sequences and modifications, activity information, etc. of different viruses such as HIV, HSV, RSV, INFVA, SARS-CoV, SARS-CoV-2, HPIV, HCV, FIV, EBOV, etc. 7255 data were deduplicated, and the final non-duplicate data set contained 6784 data.

[0085] In addition, in the specific experiment, the test activity information in the data set is divided into two categories, with Activity IC50 / EC50 10uM as the threshold, Activity IC50 / EC50 > 10uM is marked as 0, that is, the inactive category. Activity IC50 / EC50 ≤ 10uM is marked as 1, that is, the active category. It can also be divided according to multiple thresholds to convert to multi-classification prediction.

[0086] Step a2, obtain the position information corresponding to each amino acid residue in the test polypeptide sequence. The position information can be the position of each amino acid residue in the polypeptide sequence, such as the amino acid residue "A" of the sequence "AGTPIK" is located at the first position, "G" is located at the second position, and so on. Position embedding can be set as a learnable parameter, or it can be set as a sine and cosine function of different frequencies to encode each position. In this example, position embedding is set as a learnable parameter.

[0087] Step a3, based on the word-gram table, each amino acid residue in the test polypeptide sequence is segmented into word-grams, and combined with the test modification information to obtain a test word-gram set.

[0088] Step a4: Generate one-to-one embedding vectors for the test peptide samples based on the position information and the test word set. Build a deep learning model for processing sequences and modifying word units, which can be a Transformer Encoder model built from scratch. Figure 7 As shown, the input polypeptide sequence words, namely amino acid residues, modification words, special words ( <pad> 、 <cls> 、 <end>) as word embedding, plus position embedding as the input of the Transformer Encoder model. Position embedding can be set as a learnable parameter, or it can be set as a sine and cosine function of different frequencies to encode each position. In this embodiment, position embedding is set as a learnable parameter.

[0089] Step a5, input the embedded vector into the encoder structure and output the encoding result, wherein the encoder structure includes a multi-head self-attention layer, a first normalization and residual connection layer, a feedforward fully connected layer, and a second normalization and residual connection layer connected in sequence.

[0090] The embedding vector is input into the encoder, which contains multiple layers of the same structure, each of which mainly contains multi-head attention and feedforward neural networks. Residual connections are used to accelerate model training, and normalization is a regularization method to alleviate the internal covariate shift of the model.

[0091] Step a6, input the encoding result into the hidden layer, and extract the sequence tags corresponding to each test polypeptide sample.

[0092] The encoder outputs to the hidden layer and takes out <cls>The corresponding representation is usually represented by a special <cls>The tag is used to represent the aggregate information of the entire sequence, and the hidden state of this tag is extracted, which represents the comprehensive representation of the entire sequence.

[0093] Step a7, input the sequence tag into the classifier, and perform model training based on the test activity information to generate a polypeptide activity prediction model. Enter the classifier to predict the activity of the polypeptide. The classifier can be a fully connected network or a convolutional neural network. In the specific experiment, the classifier is a fully connected network, and the output of the classifier is 2, which is a binary classification problem.

[0094] In this embodiment, the non-repetitive data set contains 6784 data, the training set contains 4070 data, the validation set contains 1357 data, and the test set contains 1357 data. During model training, according to the validation set AUC (Area Under the Receiver Operating Characteristic Curve, the area under the receiver operating characteristic curve is an important indicator for evaluating the performance of the classification model) value, the optimal deep learning model is saved as the final interpretation model, that is, the polypeptide activity prediction model. Finally, the optimal model was tested in the test set, and the corresponding AUC was 0.8923.

[0095] In some optional embodiments, after generating the polypeptide activity prediction model, the following steps are included:

[0096] Step a8, loading the polypeptide activity prediction model;

[0097] Step a9, creating an interpretability framework according to the model type of the polypeptide activity prediction model to establish the interpretability sub-model of the polypeptide activity prediction model.

[0098] The interpretability framework in this embodiment is the SHAP framework, a method for explaining the prediction results of machine learning models. Based on the trained polypeptide activity prediction model, the SHAP framework is used to establish an interpretability method and calculate the SHAP value of the polypeptide activity category.

[0099] In this embodiment, it is necessary to collect polypeptide sequences and modification information and corresponding activity values, then segment the sequences and modification information into tokens, convert them into digital signals, then build and train a polypeptide activity prediction model, input the digital signals into the model for training and testing, and save the optimal activity prediction model; then build an interpretability method for the prediction model, based on the trained activity prediction model, use the SHAP framework to establish an interpretability method, calculate the SHAP value of the polypeptide sequence and modification information, explain the relationship between the polypeptide sequence and modification information and the activity value, and can be based on all SHAP values ​​of the interpretation data set, and draw a picture to show the importance of different polypeptide sequence tokens and modification tokens to the polypeptide activity type. The present invention can comprehensively analyze the influence of polypeptide sequence and modification information on polypeptide activity, and can help design active polypeptides.

[0100] In some optional embodiments, based on the interpretability sub-model of the polypeptide activity prediction model, the polypeptide activity category is used as a calculation condition to determine the activity contribution value corresponding to the amino acid residues and modification information in the polypeptide sample to be analyzed, including:

[0101] When the peptide activity prediction model is a binary classification model, the activity category is specified as a calculation condition.

[0102] When the polypeptide activity prediction model is a multi-classification model, the highest activity class or the lowest activity class is specified as the calculation condition.

[0103] The interpretability method of constructing the prediction model is to build a custom prediction function based on the SHAP framework. A category needs to be specified in the prediction function as a calculation condition. For a binary classification model, the positive category (i.e., the active category) can be specified as the calculation condition, and the SHAP value under the active condition can be calculated. For a multi-classification model, the highest activity category (i.e., the highest activity level) can be specified as the calculation condition, and the SHAP value under the highest activity condition can be calculated. For a multi-classification model, the lowest activity category (i.e., the lowest activity level, inactive) can also be specified as the calculation condition, and the SHAP value under the inactive condition can be calculated. When the multi-classification model specifies the lowest activity category as the calculation condition, it explains the positive and negative effects of the polypeptide sequence and modification when it is inactive; by changing the positive to negative and the negative to positive, the positive and negative effects of the polypeptide sequence and modification when it is active can be explained.

[0104] In this embodiment, in the binary classification model, the effect of amino acid residues on the activity category can be directly evaluated by specifying the active category. In the multi-classification model, by specifying the highest or lowest activity category, the key influencing factors of a specific category can be understood more accurately, which is conducive to improving the effectiveness of peptide design.

[0105] In some optional implementations, the word table is established by the following steps:

[0106] Define the letter abbreviations corresponding to various amino acid residues one by one to obtain a basic word unit comparison table;

[0107] Define a one-to-one correspondence between multiple modification information and special modification words, and obtain a comparison table of special modification words;

[0108] Define special tokens, which include: completion token, start token, and end token;

[0109] Define unknown tokens;

[0110] A word unit table is established based on the basic word unit comparison table, the modified special word unit comparison table, the marked special word unit, and the unknown word unit.

[0111] In this embodiment, the word list includes 20 amino acid residue letter abbreviations (Y, A, T, V, L, D, E, G, R, H, I, W, Q, K, M, F, N, S, P, C), 2 modification special words (modification 1: ~, modification 2: !), 3 other special words ( <pad> 、 <cls> 、 <end>), and 1 unknown word ( <unknown>). Among them, the three other special words <pad>Indicates padding. When the input length is less than the maximum input length, use <pad>Pad to the maximum input length. <cls>Indicates the start of input. <end>Indicates the end of input. In addition, when building the model, you can build a word table from scratch based on the word units that appear, map all the word units that appear to numbers, map the 26 word units to numbers, and establish a one-to-one correspondence between word units and numbers from scratch, that is, build a digital word table. Figure 8 shown.

[0112] In this embodiment, a word element table can be constructed from scratch based on the appearing word elements, and all the appearing word elements can be mapped to numbers; or an existing word element table can be directly used. The existing word element table needs to include all the appearing word elements in the data set, and the number of word elements in the existing word element table can be greater than the number of all the appearing word elements in the data set.

[0113] In addition, the number of tokens should be an even number and divisible by the batch size to accelerate model training and inference. When the number of tokens is an odd number, it should be padded with false tokens to an even number that is divisible by the batch size.

[0114] In this embodiment, the amino acid residues, modification information, and special words can be standardized into a unified representation through the word table, thereby improving the efficiency and consistency of model processing.

[0115] In this embodiment, a device for interpretability analysis of polypeptide activity is also provided, which is used to implement the above-mentioned embodiments and preferred embodiments, and will not be repeated hereafter. As used below, the term "module" can implement a combination of software and / or hardware of a predetermined function. Although the device described in the following embodiments is preferably implemented in software, the implementation of hardware, or a combination of software and hardware, is also possible and contemplated.

[0116] This embodiment provides an interpretable analysis device for polypeptide activity, such as Fig. 9 As shown, the device comprises:

[0117] An acquisition module 201 is used to acquire information of polypeptides to be analyzed corresponding to the polypeptide samples to be analyzed, where the information of polypeptides to be analyzed includes polypeptide sequences and modification information;

[0118] The word unit segmentation module 202 is used to perform word unit segmentation on each amino acid residue in the polypeptide sequence based on a pre-established word unit table, and obtain a polypeptide information word unit set in combination with modification information;

[0119] Prediction module 203, used to input the polypeptide information word set into a pre-established polypeptide activity prediction model to obtain the polypeptide activity category corresponding to the polypeptide sample to be analyzed;

[0120] The analysis module 204 is used to determine the activity contribution values ​​of the amino acid residues and modification information in the polypeptide sample to be analyzed based on the interpretable sub-model of the polypeptide activity prediction model and taking the polypeptide activity category as a calculation condition.

[0121] In an optional embodiment, the device further comprises:

[0122] The visualization display module is used to determine the influencing factors to be analyzed, which include: the overall impact of the polypeptide information to be analyzed on the polypeptide activity, the maximum impact of the polypeptide information to be analyzed on the polypeptide activity, and the positive and negative impact relationship between the polypeptide information to be analyzed and the polypeptide activity; according to the influencing factors to be analyzed, the activity contribution values ​​corresponding to the amino acid residues and the modification information are calculated to determine the single activity contribution values ​​corresponding to the amino acid residues and the modification information; based on the single activity contribution values ​​corresponding to the amino acid residues and the modification information, the amino acid residues and the modification information are sorted; and the sorting results and the corresponding activity contribution values ​​are visualized.

[0123] Acquire test polypeptide data corresponding to a plurality of test polypeptide samples, wherein the test polypeptide data includes test polypeptide sequence, test modification information, and test activity information;

[0124] A model building module is used to obtain the position information corresponding to each amino acid residue in the test polypeptide sequence; based on the word list, each amino acid residue in the test polypeptide sequence is segmented into words to obtain a test word set; based on the position information and the test word set, an embedding vector corresponding to the test polypeptide sample is generated; the embedding vector is input into the encoder structure and the encoding result is output, wherein the encoder structure includes a multi-head self-attention layer, a first normalization and residual connection layer, a feedforward full connection layer, and a second normalization and residual connection layer connected in sequence; the encoding result is input into the hidden layer to extract the sequence tag corresponding to each test polypeptide sample; the sequence tag is input into the classifier, and the model is trained based on the test activity information to generate a polypeptide activity prediction model; the polypeptide activity prediction model is loaded; according to the model type of the polypeptide activity prediction model, an interpretability framework is created to establish an interpretability sub-model of the polypeptide activity prediction model.

[0125] The word unit establishment module is used to define the letter abbreviations corresponding to multiple amino acid residues one by one to obtain a basic word unit comparison table; define the modified special word units corresponding to multiple modification information one by one to obtain a modified special word unit comparison table; define the mark special word units, the mark special word units include: completion mark word units, start mark word units, and end mark word units; define unknown word units; map the basic word unit comparison table, the modified special word unit comparison table, the mark special word units, and the unknown word units to numbers to establish a word unit table.

[0126] The interpretability analysis device for polypeptide activity in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0127] The further functional description of each of the above modules and units is the same as that of the above corresponding embodiments and will not be repeated here.

[0128] The embodiment of the present invention also provides a computer device having the above Fig. 9 Interpretable analysis setup for the activity of the peptides shown.

[0129] See also Fig.10 , Fig.10 is a schematic diagram of the structure of a computer device provided by an optional embodiment of the present invention, such as Fig.10 As shown, the computer device includes: one or more processors 10, a memory 20, and interfaces for connecting various components, including high-speed interfaces and low-speed interfaces. Various components are connected to each other using different buses for communication, and can be installed on a common mainboard or installed in other ways as needed. The processor can process the instructions executed in the computer device, including instructions stored in or on the memory to display the graphical information of the GUI on an external input / output device (such as, a display device coupled to the interface). In some optional embodiments, if necessary, multiple processors and / or multiple buses can be used together with multiple memories and multiple memories. Similarly, multiple computer devices can be connected, and each device provides some necessary operations (for example, as a server array, a group of blade servers, or a multi-processor system). Fig.10 A processor 10 is taken as an example.

[0130] The processor 10 may be a central processing unit, a network processor or a combination thereof. The processor 10 may further include a hardware chip. The hardware chip may be a dedicated integrated circuit, a programmable logic device or a combination thereof. The programmable logic device may be a complex programmable logic device, a field programmable gate array, a general purpose array logic or any combination thereof.

[0131] The memory 20 stores instructions executable by at least one processor 10, so that the at least one processor 10 executes the method shown in the above embodiment.

[0132] The memory 20 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the computer device, etc. In addition, the memory 20 may include a high-speed random access memory, and may also include a non-transient memory, such as at least one disk storage device, a flash memory device, or other non-transient solid-state storage device. In some optional embodiments, the memory 20 may optionally include a memory remotely arranged relative to the processor 10, and these remote memories may be connected to the computer device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0133] The memory 20 may include a volatile memory, such as a random access memory; the memory may also include a non-volatile memory, such as a flash memory, a hard disk or a solid state drive; the memory 20 may also include a combination of the above types of memory.

[0134] The computer device further comprises a communication interface 30 for the computer device to communicate with other devices or a communication network.

[0135] The embodiment of the present invention also provides a computer-readable storage medium. The method according to the embodiment of the present invention can be implemented in hardware, firmware, or can be implemented as a computer code that can be recorded in a storage medium, or can be implemented as a computer code that is originally stored in a remote storage medium or a non-temporary machine-readable storage medium and will be stored in a local storage medium through a network download, so that the method described herein can be stored in such software processing on a storage medium using a general-purpose computer, a dedicated processor, or programmable or dedicated hardware. Among them, the storage medium can be a magnetic disk, an optical disk, a read-only storage memory, a random access memory, a flash memory, a hard disk or a solid-state hard disk, etc.; further, the storage medium can also include a combination of the above types of memories. It can be understood that a computer, a processor, a microprocessor controller, or programmable hardware includes a storage component that can store or receive software or computer code. When the software or computer code is accessed and executed by a computer, a processor, or hardware, the method shown in the above embodiment is implemented.

[0136] A part of the present invention may be applied as a computer program product, such as a computer program instruction, which, when executed by a computer, can call or provide the method and / or technical solution according to the present invention through the operation of the computer. Those skilled in the art should understand that the existence of the computer program instruction in a computer-readable medium includes, but is not limited to, a source file, an executable file, an installation package file, etc., and accordingly, the way in which the computer program instruction is executed by the computer includes, but is not limited to: the computer directly executes the instruction, or the computer compiles the instruction and then executes the corresponding compiled program, or the computer reads and executes the instruction, or the computer reads and installs the instruction and then executes the corresponding installed program. Here, the computer-readable medium may be any available computer-readable storage medium or communication medium accessible to the computer.

[0137] Although the embodiments of the present invention have been described in conjunction with the accompanying drawings, those skilled in the art may make various modifications and variations without departing from the spirit and scope of the present invention, and such modifications and variations are all within the scope defined by the appended claims.< / end> < / cls> < / pad> < / pad> < / unknown> < / end> < / cls> < / pad> < / cls> < / cls> < / end> < / cls> < / pad> < / end> < / cls> < / end> < / cls> < / end> < / cls>

Claims

1. A method for interpretability analysis of polypeptide activity, characterized in that: The method comprises: Acquire the polypeptide information to be analyzed corresponding to the polypeptide sample to be analyzed, wherein the polypeptide information to be analyzed includes polypeptide sequence and modification information; Based on a pre-established word-unit table and the polypeptide information to be analyzed, each amino acid residue in the polypeptide sequence is segmented into words to obtain a polypeptide information word-unit set; Inputting the polypeptide information word set into a pre-established polypeptide activity prediction model to obtain the polypeptide activity category corresponding to the polypeptide sample to be analyzed; Based on the interpretable sub-model of the polypeptide activity prediction model, the polypeptide activity category is used as a calculation condition to determine the activity contribution value of the amino acid residues and the modification information in the polypeptide sample to be analyzed, which includes: when the polypeptide activity prediction model is a binary classification model, specifying the active category as the calculation condition; when the polypeptide activity prediction model is a multi-classification model, specifying the highest activity category or the lowest activity category as the calculation condition; The polypeptide activity prediction model is established by the following steps: Acquire test polypeptide data corresponding to a plurality of test polypeptide samples, wherein the test polypeptide data includes test polypeptide sequence, test modification information, and test activity information; Obtaining position information corresponding to each amino acid residue in the test polypeptide sequence; Based on the word-gram table, each amino acid residue in the test polypeptide sequence is segmented into words to obtain a test word-gram set; Based on the position information and the test word set, generating an embedding vector corresponding to each of the test polypeptide samples; Inputting the embedding vector into an encoder structure and outputting an encoding result, wherein the encoder structure includes a multi-head self-attention layer, a first normalization and residual connection layer, a feedforward fully connected layer, and a second normalization and residual connection layer connected in sequence; Input the encoding result into the hidden layer, and extract the sequence tag corresponding to each of the test polypeptide samples; Inputting the sequence tags into a classifier, and performing model training based on the test activity information to generate the polypeptide activity prediction model; After generating the polypeptide activity prediction model, the method comprises: Loading the polypeptide activity prediction model; According to the model type of the polypeptide activity prediction model, an interpretability framework is created to establish the interpretability sub-model of the polypeptide activity prediction model.

2. The method according to claim 1, characterized in that There are multiple activity contribution values ​​corresponding to the amino acid residues and the modification information. After determining the activity contribution values ​​corresponding to the amino acid residues and the modification information in the polypeptide sample to be analyzed, the method further includes: Determine the influencing factors to be analyzed, wherein the influencing factors to be analyzed include: the overall influence of the polypeptide information to be analyzed on the polypeptide activity, the maximum influence of the polypeptide information to be analyzed on the polypeptide activity, and the positive and negative influence relationship between the polypeptide information to be analyzed and the polypeptide activity; According to the influencing factors to be analyzed, the activity contribution values ​​corresponding to the amino acid residues and the modification information are calculated to determine a single activity contribution value corresponding to the amino acid residues and the modification information; sorting the amino acid residues and the modification information based on the single activity contribution values ​​corresponding one to one between the amino acid residues and the modification information; The sorting results and the corresponding activity contribution values ​​are visualized.

3. The method according to claim 1, characterized in that The word table is established by the following steps: Define the letter abbreviations corresponding to various amino acid residues one by one to obtain a basic word unit comparison table; Define a one-to-one correspondence between multiple modification information and special modification words, and obtain a comparison table of special modification words; Define special marker words, the special marker words include: completion marker words, start marker words, and end marker words; Define unknown tokens; The basic word unit comparison table, the modified special word unit comparison table, the marked special word unit, and the unknown word unit are mapped into numbers to establish the word unit table.

4. A device for interpretable analysis of polypeptide activity, characterized in that: The device comprises: An acquisition module is used to acquire the polypeptide information to be analyzed corresponding to the polypeptide sample to be analyzed, wherein the polypeptide information to be analyzed includes polypeptide sequence and modification information; A word unit segmentation module, used to perform word unit segmentation on each amino acid residue in the polypeptide sequence based on a pre-established word unit table, and obtain a polypeptide information word unit set in combination with the modification information; A prediction module, used to input the polypeptide information word set into a pre-established polypeptide activity prediction model to obtain the polypeptide activity category corresponding to the polypeptide sample to be analyzed; An analysis module is used to determine the activity contribution values ​​of the amino acid residues and the modification information in the polypeptide sample to be analyzed, based on the interpretable sub-model of the polypeptide activity prediction model, taking the polypeptide activity category as a calculation condition; wherein, it includes: when the polypeptide activity prediction model is a binary classification model, specifying the active category as the calculation condition; when the polypeptide activity prediction model is a multi-classification model, specifying the highest activity category or the lowest activity category as the calculation condition; A model building module is used to obtain test polypeptide data corresponding to multiple test polypeptide samples one by one, wherein the test polypeptide data includes a test polypeptide sequence, test modification information, and test activity information; obtain the position information corresponding to each amino acid residue in the test polypeptide sequence; based on the word-gram table, perform word-gram segmentation on each amino acid residue in the test polypeptide sequence to obtain a test word-gram set; based on the position information and the test word-gram set, generate an embedding vector corresponding to the test polypeptide sample one by one; input the embedding vector into an encoder structure and output an encoding result, wherein the encoder structure includes a multi-head self-attention layer, a first normalization and residual connection layer, a feedforward full connection layer, and a second normalization and residual connection layer connected in sequence; input the encoding result into a hidden layer to extract a sequence tag corresponding to each test polypeptide sample one by one; input the sequence tag into a classifier, and perform model training based on the test activity information to generate the polypeptide activity prediction model; load the polypeptide activity prediction model; create an interpretability framework according to the model type of the polypeptide activity prediction model to establish the interpretability sub-model of the polypeptide activity prediction model.

5. A computer device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the interpretability analysis method of polypeptide activity according to any one of claims 1 to 3 by executing the computer instructions.

6. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores computer instructions, and the computer instructions are used to enable the computer to execute the interpretable analysis method of polypeptide activity according to any one of claims 1 to 3.

7. A computer program product, characterized in that The invention comprises computer instructions, wherein the computer instructions are used to make a computer execute the interpretable analysis method of the polypeptide activity according to any one of claims 1 to 3.

Images