Polypeptide generation method, device and computer equipment

Abstract

The present invention relates to the field of bioinformatics, and discloses a polypeptide generation method, device and computer equipment, the method comprising: generating a first polypeptide based on a polypeptide generation model; using a polypeptide activity prediction model to predict the activity information of the first polypeptide; using a polypeptide toxicity prediction model to predict the toxicity information of the first polypeptide; performing interpretability analysis on the polypeptide activity prediction model and the polypeptide toxicity prediction model respectively; adjusting the sequence of the second polypeptide based on the interpretability analysis result; the second polypeptide is the first polypeptide predicted to be active by the polypeptide activity prediction model and predicted to be non-toxic by the polypeptide toxicity prediction model. The present invention can greatly shorten the polypeptide drug research and development cycle, reduce research and development costs, and improve the quality of polypeptide drug research and development.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics, and in particular to a polypeptide generation method, device and computer equipment. Background Art

[0002] Peptides have antiviral, antibacterial, and anticancer properties, and can cross the blood-brain barrier, making them a hot topic in drug development. The development process of peptide drugs includes multiple stages, among which activity screening and toxicity screening are very important links. These screening processes ensure that candidate drugs are both effective and safe and suitable for entering the clinical trial stage. The purpose of activity screening is to evaluate the effectiveness of candidate peptide drugs against target diseases. This requires preliminary efficacy testing under laboratory conditions using models such as cell lines or enzymes. For example, receptor binding experiments can be used to test whether the peptide can effectively bind to the target receptor. Toxicity screening is to ensure the safety of candidate drugs and avoid potential side effects. Toxicity screening involves evaluating the toxicity of drugs at the cellular level, which can be performed through cell viability assays such as MTT (diphenyltetrazolium bromide) and WST-1 (a compound similar to MTT).

[0003] At present, the activity screening and toxicity screening in the peptide drug development process are mainly completed through experiments, which has a long development cycle and high cost. Summary of the invention

[0004] In view of this, the present invention provides a polypeptide generation method, device and computer equipment to solve the problems of long cycle and high cost in the research and development of polypeptide drugs.

[0005] In a first aspect, the present invention provides a method for producing a polypeptide, the method comprising:

[0006] generating a first polypeptide based on the polypeptide generation model;

[0007] Using a polypeptide activity prediction model, based on the sequence information and structural information of the first polypeptide and the sequence information and structural information of the target receptor, the activity information of the first polypeptide is predicted;

[0008] Using a polypeptide toxicity prediction model, based on the sequence information, structural information and composition information of the first polypeptide, and the sequence information, structural information and composition information of the target receptor, predict the toxicity information of the first polypeptide;

[0009] Performing interpretability analysis on the polypeptide activity prediction model and the polypeptide toxicity prediction model respectively;

[0010] Based on the interpretability analysis results, the sequence of the second polypeptide is adjusted; the second polypeptide is the first polypeptide predicted to be active by the polypeptide activity prediction model and predicted to be non-toxic by the polypeptide toxicity prediction model.

[0011] In an optional embodiment, adjusting the sequence of the second polypeptide based on the interpretability analysis result comprises:

[0012] Deleting a first word satisfying a first preset condition from the second polypeptide, wherein the first preset condition is that the contribution to toxicity is a positive value and the contribution to activity is a negative value;

[0013] and / or,

[0014] Using a second word element that meets a second preset condition to replace the first word element in the second polypeptide; the second preset condition is that the contribution to toxicity is a negative value and the contribution to activity is a positive value;

[0015] and / or,

[0016] The second word is used to splice into the second polypeptide.

[0017] In an optional embodiment, after adjusting the sequence of the second polypeptide based on the interpretability analysis result, the method further comprises:

[0018] obtaining the adjusted third polypeptide;

[0019] Using a polypeptide activity prediction model, based on the sequence information and structural information of the third polypeptide and the sequence information and structural information of the target receptor, the activity information of the third polypeptide is obtained;

[0020] The toxicity information of the third polypeptide is predicted using a polypeptide toxicity prediction model based on the sequence information, structural information and composition information of the third polypeptide and the sequence information, structural information and composition information of the target receptor.

[0021] In an optional embodiment, the polypeptide generation model includes one or more Transformer decoders, layer normalization, linear layers and softmax layers connected in sequence;

[0022] The Transformer decoder includes a multi-head self-attention layer, a residual connection, a layer normalization, and a feed-forward network layer.

[0023] In an optional embodiment, the polypeptide activity prediction model includes a first Star-Transformer model, two first visual Transformer models, a first Transformer encoder, and multiple multi-layer perceptrons;

[0024] The first Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor;

[0025] The two first visual Transformer models are used to extract features from the structural information of the first polypeptide and the structural information of the target receptor respectively; the feature information extracted by the two first visual Transformer models respectively enters the first Transformer encoder through a multi-layer perceptron for feature fusion;

[0026] The outputs of the first Star-Transformer model and the first Transformer encoder are fused after passing through a multi-layer perceptron respectively, and the fused features enter the classifier built based on the multi-layer perceptron for prediction to obtain the activity information of the first polypeptide.

[0027] In an optional embodiment, the input of the first visual Transformer model for extracting features of the structural information of the target receptor is a three-dimensional structural image block of the target receptor, and the input of the first visual Transformer model for extracting features of the structural information of the first polypeptide is a three-dimensional structural image block of the first polypeptide;

[0028] The size of the target receptor three-dimensional structure image block is larger than the size of the first polypeptide three-dimensional structure image block.

[0029] In an optional embodiment, the polypeptide toxicity prediction model includes a second Star-Transformer model, two second visual Transformer models, a third visual Transformer model, a second Transformer encoder, and multiple multi-layer perceptrons;

[0030] The second Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor;

[0031] The two second visual Transformer models are used to extract features from the structural information of the first polypeptide and the structural information of the target receptor respectively; the feature information extracted by the two second visual Transformer models respectively enters the second Transformer encoder through a multi-layer perceptron for feature fusion;

[0032] The third visual Transformer model is used to extract features from the composition information of the first polypeptide and the composition information of the target receptor;

[0033] The outputs of the second Star-Transformer model, the second Transformer encoder, and the third visual Transformer model are respectively fused after passing through a multi-layer perceptron, and the fused features enter the classifier built based on the multi-layer perceptron for prediction to obtain the toxicity information of the first polypeptide.

[0034] In an optional embodiment, the input of the second visual Transformer model for extracting features of the structural information of the target receptor is a three-dimensional structural image block of the target receptor, and the input of the second visual Transformer model for extracting features of the structural information of the first polypeptide is a three-dimensional structural image block of the first polypeptide;

[0035] The size of the target receptor three-dimensional structure image block is larger than the size of the first polypeptide three-dimensional structure image block.

[0036] In an optional embodiment, after generating the first polypeptide based on the polypeptide generation model, the method further comprises:

[0037] Using an artificial intelligence model to predict the sequence information of the first polypeptide to obtain a structure prediction result;

[0038] Based on the structure prediction result, a structural image of the first polypeptide is obtained as the structural information of the first polypeptide.

[0039] In a second aspect, the present invention provides a polypeptide production device, the device comprising:

[0040] A generation module, used to generate a first polypeptide based on a polypeptide generation model;

[0041] An activity prediction module, used to obtain the activity information of the first polypeptide based on the sequence information and structural information of the first polypeptide and the sequence information and structural information of the target receptor using a polypeptide activity prediction model;

[0042] A toxicity prediction module, used to predict the toxicity information of the first polypeptide based on the sequence information, structure information and composition information of the first polypeptide and the sequence information, structure information and composition information of the target receptor using a polypeptide toxicity prediction model;

[0043] An explanation module, used to perform interpretability analysis on the polypeptide activity prediction model and the polypeptide toxicity prediction model respectively;

[0044] An adjustment module is used to adjust the sequence of a second polypeptide based on the interpretability analysis result; the second polypeptide is the first polypeptide predicted to be active by the polypeptide activity prediction model and predicted to be non-toxic by the polypeptide toxicity prediction model.

[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 polypeptide generation method of the first aspect or any corresponding embodiment thereof by executing the computer instructions.

[0046] In a fourth aspect, the present invention provides a computer-readable storage medium having computer instructions stored thereon, the computer instructions being used to cause a computer to execute the polypeptide generation method 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, wherein the computer instructions are used to enable a computer to execute the polypeptide generation method of the first aspect or any corresponding embodiment thereof.

[0048] The polypeptide generation method, device and computer equipment provided in this embodiment include four processes: polypeptide generation, polypeptide activity screening, polypeptide toxicity screening and polypeptide sequence adjustment, thereby establishing a complete set of methods from scratch design to screening of activity toxicity indicators and adjustment and optimization of polypeptide sequences. Among them, polypeptide generation, polypeptide activity screening and polypeptide toxicity screening are all realized by using artificial intelligence models, which reduces the demand for experiments, can greatly shorten the polypeptide drug development cycle, reduce research and development costs, and can also improve the quality of polypeptide drug development. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments 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 method for producing a polypeptide according to an embodiment of the present invention;

[0051] Figure 2 is a distribution diagram of data sets used for peptide generation model training and word library construction according to an embodiment of the present invention;

[0052] Figure 3 is a partial schematic diagram of a polypeptide word library according to an embodiment of the present invention;

[0053] Figure 4 is a box plot of the length compression ratio of converting a polypeptide sequence into a digital sequence using a word library according to an embodiment of the present invention;

[0054] Figure 5 is a schematic diagram of the structure of a multimodal activity prediction model according to an embodiment of the present invention;

[0055] Figure 6 is a schematic diagram of a process of extracting visual structure features according to an embodiment of the present invention;

[0056] Figure 7 is a schematic diagram of the structure of a multimodal toxicity prediction model according to an embodiment of the present invention;

[0057] Figure 8 is a schematic diagram of the distribution of a training data set of a multimodal toxicity prediction model according to an embodiment of the present invention;

[0058] Fig. 9 is a structural block diagram of a polypeptide generating device 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 generating a polypeptide 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 executable computer instructions, and although a logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in an order different from that shown here.

[0062] First, it should be noted that, for ease of description, the sequence of the polypeptide in the embodiments of the present invention may include amino acid residues and modifications.

[0063] In this embodiment, a method for generating a polypeptide is provided, which can be used in a computer device. Figure 1 is a flow chart of a method for generating a polypeptide according to an embodiment of the present invention, such as Figure 1 As shown, the process includes the following steps:

[0064] Step S101, generating a first polypeptide based on a polypeptide generation model.

[0065] In the embodiment of the present invention, the polypeptide generation model generates the first polypeptide one word at a time, and the generated words are arranged in the generation order to obtain the sequence of the first polypeptide. Specifically, the polypeptide generation model outputs a set of probability values ​​each time, each probability value in a set of probability values ​​corresponds to a word in the word library, and the word corresponding to the largest probability value in the set of probability values ​​is the currently generated word.

[0066] The above database can be constructed in the following way: obtain a peptide data set, convert the amino acid residues and modifications of the peptides in the data set into numbers, and then build a word library according to the appearance rules of the numbers corresponding to the amino acid residue combinations and modification combinations in the existing peptide data.

[0067] Specifically, a total of 1657 non-repetitive polypeptides can be obtained as a data set from the DRAVP database (http: / / dravp.cpu-bioinfor.org / ), the AVPdb database (http: / / crdd.osdd.net / servers / avpdb / ), and other places. Figure 2 As shown, the dataset contains peptide data of multiple viruses, namely, peptides of human immunodeficiency virus (HIV), respiratory syncytial virus (RSV), influenza virus (INFV), hepatitis C virus (HCV), severe acute respiratory syndrome coronavirus (SARS-COV), new coronavirus (SARS-COV-2), feline immunodeficiency virus (FIV), herpes simplex virus (HSV), and human parainfluenza virus (HPIV).

[0068] In the word library, it is necessary to set the longest word length in the word library, and discard words that exceed the longest word length. The longest word length should be less than the shortest length of the existing polypeptide sequence, that is, each polypeptide sequence is converted into at least two words. The minimum length of all words in the word library is 1, which is the length represented by a single amino acid residue letter. It is necessary to set the maximum number of all words contained in the word library to avoid too many words in the word library and too many rare words. This will cause the peptide generation model to generate too many rare words, affecting the quality of polypeptide design. When building a word library, it is also necessary to set the minimum number of occurrences of each word. Words with less than the minimum number of occurrences will be regarded as rare words and discarded. The word library should include special words such as separation, completion, and unknown. For example Figure 3As shown in the figure, the constructed word library includes 20 amino acid residue abbreviations (A, R, N, D, C, Q, E, G, H, I, L, K, M, F, P, S, T, W, Y, V), newly generated combined word elements (EE, KK, IEE, VN, etc.), modified word elements ( <m1> 、 <m2> 、 <m3>)、Special words( <pad> 、 <sep> 、 <unknown>)wait.

[0069] In addition, in the embodiment of the present invention, the polypeptides with IC50 / EC50<1μM in the above data set can be screened out as active polypeptides, and the polypeptides with IC50 / EC50≥1μM can be marked as inactive polypeptides, and the active polypeptides are used as training samples for the polypeptide generation model. The main training process of the polypeptide generation model is: first, the sequence of the receptor corresponding to the active polypeptide is input into the polypeptide generation model to be trained to obtain the first prediction word, and then the first word of the active polypeptide is input into the polypeptide generation model to be trained to predict the second prediction word, and then the first word and the second word of the active polypeptide are input into the polypeptide generation model to be trained to predict the third prediction word, and so on, each time the 1st to Nth word of the active polypeptide is input into the polypeptide generation model to be trained to obtain the N+1th prediction word. The number of predicted word units finally obtained is the same as the number of word units of the active polypeptide. The word unit prediction accuracy of the polypeptide generation model to be trained is calculated, and the parameters of the polypeptide generation model to be trained are adjusted based on the prediction accuracy.

[0070] The word library also includes the numbers corresponding to each word, based on which the polypeptide sequence can be converted into a digital sequence. Figure 4 As shown, compared with directly using the 20 single-letter abbreviations of amino acid residues as word units to convert polypeptide sequences into digital sequences, the length of the digital sequences decreased in different proportions based on the constructed word unit library, and the median length decrease was 2.73 times.

[0071] In the embodiment of the present invention, the input of the peptide generation model is the sequence of the target receptor to generate a peptide that binds to the target receptor. Here, the sequence of the target receptor can be directly converted into a digital sequence using the above-mentioned peptide word library, and no separate word library is needed.

[0072] In some optional embodiments, the polypeptide generation model includes one or more Transformer decoders, layer normalization, linear layers, and softmax layers connected in sequence;

[0073] The Transformer decoder includes a multi-head self-attention layer, a residual connection, a layer normalization, and a feed-forward network layer.

[0074] In the embodiment of the present invention, the peptide generation model is constructed using the structure of the Transformer decoder, which includes a multi-head self-attention layer, a residual connection, a layer normalization, a feedforward network layer and other structures in sequence. The layer normalization can be before the multi-head self-attention layer, and the layer normalization can also be before the feedforward network layer. The Transformer decoder can be repeated multiple times. The Transformer decoder is followed by a layer normalization, a linear layer, and a softmax layer, and the output size is the number of words in the peptide word library.

[0075] After the first polypeptide is generated based on the polypeptide generation model, the method further comprises:

[0076] Using an artificial intelligence model to predict the sequence information of the first polypeptide to obtain a structure prediction result;

[0077] Based on the structure prediction result, a structural image of the first polypeptide is obtained as the structural information of the first polypeptide.

[0078] In the embodiment of the present invention, the structure of the first polypeptide generated by the above polypeptide generation model can be predicted using an artificial intelligence method, and the prediction result can be converted into a picture using the pymol-open-source tool. In the following, the structural information of the target receptor is needed to predict the toxicity and activity of the first polypeptide. If the structure of the target receptor is unknown, it can also be predicted using an artificial intelligence method, and the prediction result can also be converted into a picture using the pymol-open-source tool.

[0079] The peptide activity prediction model and peptide toxicity prediction model mentioned below need to be based on the sequence information of the first peptide and the receptor for prediction. Therefore, a word library is needed to convert the sequence information into digital information. The peptide sequence adjustment stage also requires a word library to convert the sequence information into digital information. The peptide word library for peptide activity screening, peptide toxicity screening, and peptide sequence adjustment can directly use the word library of the peptide generation stage, or it can use new word library construction methods such as kmer to rebuild the word library. The specific word library to be used needs to be evaluated according to the model performance indicators in the above links, and then the word library corresponding to the optimal model performance is selected.

[0080] Step S102, using a polypeptide activity prediction model, based on the sequence information and structural information of the first polypeptide, and the sequence information and structural information of the target receptor, predict the activity information of the first polypeptide.

[0081] In some specific implementations, such as Figure 5 As shown, the peptide activity prediction model includes a first Star-Transformer model, two first visual Transformer models, a first Transformer encoder, and multiple multi-layer perceptrons (MLPs);

[0082] The first Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor;

[0083] The two first visual Transformer models are used to extract features from the structural information of the first polypeptide and the structural information of the target receptor respectively; the feature information extracted by the two first visual Transformer models respectively enters the first Transformer encoder through a multi-layer perceptron for feature fusion;

[0084] The outputs of the first Star-Transformer model and the first Transformer encoder are fused after passing through a multi-layer perceptron respectively, and the fused features enter the classifier built based on the multi-layer perceptron for prediction to obtain the activity information of the first polypeptide.

[0085] In the embodiment of the present invention, the peptide activity prediction model integrates the sequence features of the peptide and the receptor, and the structural features of the peptide and the receptor. The sequences of the receptor and the first peptide are directly input into a Star-Transformer model for sequence feature extraction, and the model is connected to a fully connected layer, that is, Figure 5 The multi-layer perceptron (MLP) layer in the image processing unit. The structural features of the first peptide and the receptor are extracted using two different vision transformer models, and the vision transformer model is connected to the feature fusion model transformer encoder.

[0086] like Figure 6 As shown in the figure, the peptide structure image is cut into 8 image blocks, which are similar to the 8 words in natural language processing. Then it is flattened linearly projected, and the special word 0 is added. Finally, it is input into the Transformer encoder together with the position code. The Transformer encoder is connected to the fully connected layer.

[0087] In an optional embodiment, the input of the first visual Transformer model for extracting features of the structural information of the target receptor is a three-dimensional (3D) structural image block of the target receptor, and the input of the first visual Transformer model for extracting features of the structural information of the first polypeptide is a first polypeptide three-dimensional structural image block;

[0088] The size of the target receptor three-dimensional structure image block is larger than the size of the first polypeptide three-dimensional structure image block.

[0089] In an embodiment of the present invention, an asymmetric visual feature extraction model (i.e., a visual Transformer model) is used to extract visual structural features of the first polypeptide and the receptor, respectively. Specifically, two different models are used to extract visual structural features of the first polypeptide and the receptor, respectively. Since the receptor structure is usually more complex than the polypeptide structure, the size of the receptor structural feature extraction model is larger than the size of the polypeptide structural feature extraction model. The sizes of the two models are asymmetric, which is convenient for extracting different features. The receptor visual feature extraction model and the polypeptide visual feature extraction model can both use a visual Transformer model, and the input is a 3D structural image. The size of the receptor 3D structural image is consistent with the size of the polypeptide 3D structural image. The size of the input image block of the receptor visual feature extraction model is asymmetric with the input image block of the polypeptide visual feature extraction model. The size of the input image block of the receptor visual feature extraction model (i.e., the target receptor three-dimensional structure image block) is larger than the size of the input image block of the polypeptide visual feature extraction model (i.e., the first polypeptide three-dimensional structure image block). Accordingly, the number of input image blocks of the receptor visual feature extraction model is less than the number of input image blocks of the polypeptide visual feature extraction model. The features extracted by the receptor visual feature extraction model and the features extracted by the polypeptide visual feature extraction model are of the same size, which facilitates feature alignment and fusion. The Transformer encoder model can be used to align and fuse the two extracted visual features. The Transformer encoder structure can be repeated multiple times for deep fusion of the two extracted visual features.

[0090] In other embodiments, the asymmetric visual feature extraction model may not be connected to a feature fusion model, and the features extracted by the receptor visual feature extraction model and the features extracted by the polypeptide visual feature extraction model may be directly spliced ​​or added. At this time, the size of the features extracted by the receptor visual feature extraction model and the features extracted by the polypeptide visual feature extraction model may be consistent or inconsistent, which does not affect the feature splicing or addition operation. Regardless of whether the Transformer encoder model is used or not, the visual receptor visual feature extraction model and the polypeptide visual feature extraction model are followed by a fully connected layer.

[0091] The polypeptides in the polypeptide activity prediction model data set provided in the embodiment of the present invention are used as training samples.

[0092] Step S103, using a polypeptide toxicity prediction model, based on the sequence information, structure information and composition information of the first polypeptide, and the sequence information, structure information and composition information of the target receptor, predict the toxicity information of the first polypeptide.

[0093] In some specific implementations, such as Figure 7 As shown, the peptide toxicity prediction model includes a second Star-Transformer model, two second visual Transformer models, a third visual Transformer model, a second Transformer encoder, and multiple multi-layer perceptrons;

[0094] The second Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor;

[0095] The two second visual Transformer models are used to extract features from the structural information of the first polypeptide and the structural information of the target receptor respectively; the feature information extracted by the two second visual Transformer models respectively enters the second Transformer encoder through a multi-layer perceptron for feature fusion;

[0096] The third visual Transformer model is used to extract features from the composition information of the first polypeptide and the composition information of the target receptor;

[0097] The outputs of the second Star-Transformer model, the second Transformer encoder, and the third visual Transformer model are respectively fused after passing through a multi-layer perceptron, and the fused features enter the classifier built based on the multi-layer perceptron for prediction to obtain the toxicity information of the first polypeptide.

[0098] Like the above-mentioned polypeptide activity prediction model, the embodiment of the present invention also uses an asymmetric visual feature extraction model (i.e., a visual Transformer model) to extract visual structural features of the first polypeptide and the receptor respectively. The input of the second visual Transformer model for extracting the structural information of the target receptor is the three-dimensional structural image block of the target receptor, and the input of the second visual Transformer model for extracting the structural information of the first polypeptide is the three-dimensional structural image block of the first polypeptide; the size of the three-dimensional structural image block of the target receptor is larger than the size of the three-dimensional structural image block of the first polypeptide.

[0099] In an embodiment of the present invention, the sequences of the receptor and the first polypeptide are input into a Star-Transformer model for sequence feature extraction, and a fully connected layer is connected to the model. The visual structure image of the receptor and the visual structure image of the polypeptide are respectively input into two visual Transformer models of asymmetric sizes, and two outputs are obtained through two fully connected layers respectively, which are merged and input into the Transformer encoder for feature fusion, and finally the output of the fully connected layer is obtained. The receptor composition information and the polypeptide composition information are input into a one-dimensional visual Transformer model for composition feature extraction, and then the output of the fully connected layer is obtained. Finally, the fully connected layer outputs from the three models, namely the Star-Transformer model, the visual Transformer model + Transformer encoder model, and the one-dimensional visual Transformer model, are spliced ​​and input into the classifier to predict the toxicity of the polypeptide.

[0100] The training samples of the polypeptide toxicity prediction model provided in the embodiment of the present invention can be 713 non-repetitive polypeptides, such as Figure 8 As shown in the figure, these peptides are mainly RSV and INFV peptide data, and other viruses include HIV, HCV, SARS-COV, SARS-COV-2, FIV, HSV, HPIV, etc. Data with CC50 ≤40 nM are marked as 1, which means toxic. Conversely, CC50>40nM means non-toxic.

[0101] In the embodiment of the present invention, the activity screening can be placed before the toxicity screening, and the activity of the first polypeptide generated in large quantities can be screened first, and then the toxicity screening can be performed. It is also possible to perform toxicity screening first, and then activity screening. It is necessary to set the prediction probability threshold of the polypeptide activity prediction model and the polypeptide toxicity prediction model to screen the polypeptides. Only when the prediction probability is higher than the set prediction probability threshold (the higher the prediction probability of the polypeptide activity prediction model, the higher the activity; the higher the prediction probability of the polypeptide toxicity prediction model, the lower the possibility of toxicity), the corresponding polypeptide is retained. The higher the prediction probability, the smaller the uncertainty of the model prediction. For the binary classification model, the prediction probability threshold is set to 0.5 at the lowest, which means that the probability of predicting a certain activity or toxicity classification category is greater than 50%. The polypeptide activity prediction model and the polypeptide toxicity prediction model can be designed with the same prediction probability threshold, or different prediction probability thresholds can be set. The specific threshold needs to be set according to the importance of the screening index and the number of polypeptides that meet the conditions. For example, if the importance of the activity index is set to be higher than the toxicity screening index, the prediction threshold of the activity screening model needs to be set higher than the prediction threshold of the toxicity model screening. The higher the prediction probability threshold, the fewer the number of polypeptides that meet the screening criteria. When the threshold is set too high, there may be no polypeptide sequences that meet the screening criteria. The prediction probability threshold should be less than 1. In this embodiment, activity screening is placed before toxicity screening, and the prediction probability thresholds of the polypeptide activity prediction model and the polypeptide toxicity prediction model are both set to 0.5.

[0102] Step S104, performing interpretability analysis on the polypeptide activity prediction model and the polypeptide toxicity prediction model respectively.

[0103] In the embodiments of the present invention, interpretable analysis methods for polypeptide activity prediction models and polypeptide toxicity prediction models can be respectively constructed, and the polypeptide sequences of the above-mentioned polypeptides predicted to be active and non-toxic are adjusted according to the interpretable analysis results.

[0104] Specifically, considering that only the amino acid residues and modifications of the polypeptide are adjusted in the polypeptide sequence adjustment, the polypeptide activity prediction model and the polypeptide toxicity prediction model need to be rebuilt, and the input of both models is only the sequence information of the polypeptide. The interpretability analysis method of the polypeptide activity prediction model and the polypeptide toxicity prediction model needs to be able to explain the positive and negative effects of the polypeptide sequence on the activity and toxicity of the polypeptide, so as to facilitate the adjustment of the words and modifications with high toxicity contribution and low activity contribution. When building the interpretability analysis method of the polypeptide activity prediction model and the polypeptide toxicity prediction model, it is necessary to use all the data in the data set to facilitate a more comprehensive analysis of the effect of the sequence on the activity and toxicity of the polypeptide, and when displaying the analysis results, a single polypeptide can be analyzed, or all polypeptides can be analyzed as a whole. In this embodiment, the deep learning interpretation framework SHAP (SHapley Additive exPlanations) is used for interpretable analysis, and all data are used for overall analysis during the analysis process.

[0105] Step S105, based on the interpretability analysis result, adjusting the sequence of the second polypeptide; the second polypeptide is the first polypeptide predicted to be active by the polypeptide activity prediction model and predicted to be non-toxic by the polypeptide toxicity prediction model.

[0106] In some optional specific embodiments, the sequence of the second polypeptide is adjusted based on the interpretability analysis result, specifically in the following ways:

[0107] 1. Deleting a first word satisfying a first preset condition from the second polypeptide, wherein the first preset condition is that the contribution to toxicity is a positive value and the contribution to activity is a negative value.

[0108] That is, directly delete the words that contribute more to toxicity and less to activity, or directly delete the words that have a positive impact on toxicity and a negative impact on activity.

[0109] 2. Use a second word that meets a second preset condition to replace the first word in the second polypeptide; the second preset condition is that the contribution to toxicity is a negative value and the contribution to activity is a positive value.

[0110] That is, replace the word that contributes more to toxicity and less to activity with the word that contributes less to toxicity and more to activity. In other words, replace the word that has a positive impact on toxicity and a negative impact on activity with the word that has a negative impact on toxicity and a positive impact on activity.

[0111] 3. Use the second word to splice into the second polypeptide. That is, splice the word that contributes less to toxicity and more to activity into the second polypeptide. The spliced ​​second word can be one or more. The word can correspond to an amino acid residue or a modification. The amino acid residue can be directly spliced ​​to the left or right of the second polypeptide, and the modification can be added to the tail of the second polypeptide or to other places.

[0112] In the embodiments of the present invention, the above three polypeptide adjustment methods can be performed individually or in combination.

[0113] In some optional embodiments, after adjusting the sequence of the second polypeptide based on the interpretability analysis result, the method further comprises:

[0114] obtaining the adjusted third polypeptide;

[0115] Using a polypeptide activity prediction model, based on the sequence information and structural information of the third polypeptide and the sequence information and structural information of the target receptor, the activity information of the third polypeptide is obtained;

[0116] The toxicity information of the third polypeptide is predicted using a polypeptide toxicity prediction model based on the sequence information, structural information and composition information of the third polypeptide and the sequence information, structural information and composition information of the target receptor.

[0117] In the embodiment of the present invention, the adjusted polypeptide sequence needs to be input into the polypeptide activity prediction model and the polypeptide toxicity prediction model again for verification, and it is necessary to ensure that the adjusted polypeptide is still predicted to be active and non-toxic.

[0118] The polypeptide generation method provided in this embodiment includes four processes: polypeptide generation, polypeptide activity screening, polypeptide toxicity screening, and polypeptide sequence adjustment, thereby establishing a complete set of methods from scratch design to screening of activity toxicity indicators, as well as adjustment and optimization of polypeptide sequences. Among them, polypeptide generation, polypeptide activity screening, and polypeptide toxicity screening are all achieved using artificial intelligence models, which reduces the demand for experiments, can greatly shorten the polypeptide drug development cycle, reduce research and development costs, and can also improve the quality of polypeptide drug development.

[0119] In the present embodiment, a polypeptide generating device is also provided, and the device is used to implement the above-mentioned embodiment and preferred embodiment, and the descriptions thereof are omitted. 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 with software, the implementation of hardware, or a combination of software and hardware is also possible and contemplated.

[0120] This embodiment provides a polypeptide production device, such as Fig. 9 As shown, including:

[0121] A generation module 901 is used to generate a first polypeptide based on a polypeptide generation model;

[0122] An activity prediction module 902 is used to obtain the activity information of the first polypeptide based on the sequence information and structure information of the first polypeptide and the sequence information and structure information of the target receptor using a polypeptide activity prediction model;

[0123] Toxicity prediction module 903, used to predict toxicity information of the first polypeptide based on the sequence information, structure information and composition information of the first polypeptide and the sequence information, structure information and composition information of the target receptor using a polypeptide toxicity prediction model;

[0124] An explanation module 904 is used to perform interpretability analysis on the polypeptide activity prediction model and the polypeptide toxicity prediction model respectively;

[0125] The adjustment module 905 is used to adjust the sequence of the second polypeptide based on the interpretability analysis result; the second polypeptide is the first polypeptide predicted to be active by the polypeptide activity prediction model and predicted to be non-toxic by the polypeptide toxicity prediction model.

[0126] In some optional implementations, the adjustment module 905 includes:

[0127] a deleting unit, configured to delete a first word satisfying a first preset condition from the second polypeptide, wherein the first preset condition is that the contribution to toxicity is a positive value and the contribution to activity is a negative value;

[0128] and / or,

[0129] A replacement unit, used to replace the first word in the second polypeptide with a second word that meets a second preset condition; the second preset condition is that the contribution to toxicity is a negative value and the contribution to activity is a positive value;

[0130] and / or,

[0131] A splicing unit is used for splicing into the second polypeptide using the second word.

[0132] In some optional embodiments, the polypeptide generating device further comprises:

[0133] An adjusted polypeptide acquisition module is used to acquire the adjusted third polypeptide;

[0134] An activity verification module, used to obtain the activity information of the third polypeptide based on the sequence information and structural information of the third polypeptide and the sequence information and structural information of the target receptor using a polypeptide activity prediction model;

[0135] The toxicity verification module is used to predict the toxicity information of the third polypeptide based on the sequence information, structure information and composition information of the third polypeptide and the sequence information, structure information and composition information of the target receptor by using the polypeptide toxicity prediction model.

[0136] In some optional embodiments, the polypeptide generation model includes one or more Transformer decoders, layer normalization, linear layers, and softmax layers connected in sequence;

[0137] The Transformer decoder includes a multi-head self-attention layer, a residual connection, a layer normalization, and a feed-forward network layer.

[0138] In some optional embodiments, the polypeptide activity prediction model includes a first Star-Transformer model, two first visual Transformer models, a first Transformer encoder, and multiple multi-layer perceptrons;

[0139] The first Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor;

[0140] The two first visual Transformer models are used to extract features from the structural information of the first polypeptide and the structural information of the target receptor respectively; the feature information extracted by the two first visual Transformer models respectively enters the first Transformer encoder through a multi-layer perceptron for feature fusion;

[0141] The outputs of the first Star-Transformer model and the first Transformer encoder are fused after passing through a multi-layer perceptron respectively, and the fused features enter the classifier built based on the multi-layer perceptron for prediction to obtain the activity information of the first polypeptide.

[0142] In some optional embodiments, the polypeptide toxicity prediction model includes a second Star-Transformer model, two second visual Transformer models, a third visual Transformer model, a second Transformer encoder, and multiple multi-layer perceptrons;

[0143] The second Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor;

[0144] The two second visual Transformer models are used to extract features from the structural information of the first polypeptide and the structural information of the target receptor respectively; the feature information extracted by the two second visual Transformer models respectively enters the second Transformer encoder through a multi-layer perceptron for feature fusion;

[0145] The third visual Transformer model is used to extract features from the composition information of the first polypeptide and the composition information of the target receptor;

[0146] The outputs of the second Star-Transformer model, the second Transformer encoder, and the third visual Transformer model are respectively fused after passing through a multi-layer perceptron, and the fused features enter the classifier built based on the multi-layer perceptron for prediction to obtain the toxicity information of the first polypeptide.

[0147] In some optional embodiments, the polypeptide generating device further comprises:

[0148] A structure prediction module, used to predict the sequence information of the first polypeptide using an artificial intelligence model to obtain a structure prediction result;

[0149] A structural image acquisition module is used to obtain a structural image of the first polypeptide based on the structural prediction result as the structural information of the first polypeptide.

[0150] 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.

[0151] The polypeptide generating device in this embodiment is presented in the form of a functional unit, where the unit refers to an ASIC (Application Specific Integrated Circuit) circuit, a processor and memory that executes one or more software or fixed programs, and / or other devices that can provide the above functions.

[0152] The embodiment of the present invention also provides a computer device having the above Fig. 9 The polypeptide production apparatus shown.

[0153] 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.

[0154] 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.

[0155] 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.

[0156] 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.

[0157] 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.

[0158] The computer device also includes an input device 30 and an output device 40. The processor 10, the memory 20, the input device 30 and the output device 40 may be connected via a bus or other means. Fig.10 The example of connecting through bus is taken in the following.

[0159] The input device 30 can receive input digital or character information, and generate key signal input related to the user settings and function control of the computer device, such as a touch screen, a keypad, a mouse, a track pad, a touch pad, an indicator bar, one or more mouse buttons, a trackball, a joystick, etc. The output device 40 may include a display device, an auxiliary lighting device (e.g., an LED) and a tactile feedback device (e.g., a vibration motor), etc. The above-mentioned display device includes but is not limited to a liquid crystal display, a light emitting diode, a display and a plasma display. In some optional embodiments, the display device may be a touch screen.

[0160] The computer device also includes a communication interface, which is used for the computer device to communicate with other devices or a communication network.

[0161] 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.

[0162] 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.

[0163] 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.< / unknown> < / sep> < / pad> < / m2> < / m1>

Claims

1. A method for producing a polypeptide, characterized in that: The method comprises: generating a first polypeptide based on the polypeptide generation model; Using a polypeptide activity prediction model, based on the sequence information and structural information of the first polypeptide and the sequence information and structural information of the target receptor, predict the activity information of the first polypeptide; Using a polypeptide toxicity prediction model, based on the sequence information, structural information, and composition information of the first polypeptide, and the sequence information, structural information, and composition information of the target receptor, predict toxicity information of the first polypeptide; Performing interpretability analysis on the polypeptide activity prediction model and the polypeptide toxicity prediction model respectively; Based on the interpretability analysis result, adjusting the sequence of the second polypeptide; the second polypeptide is the first polypeptide predicted to be active by the polypeptide activity prediction model and predicted to be non-toxic by the polypeptide toxicity prediction model; the adjusting the sequence of the second polypeptide based on the interpretability analysis result specifically includes: Deleting a first word satisfying a first preset condition from the second polypeptide, wherein the first preset condition is that the contribution to toxicity is positive and the contribution to activity is negative; and / or replacing the first word in the second polypeptide with a second word satisfying a second preset condition, wherein the second preset condition is that the contribution to toxicity is negative and the contribution to activity is positive; and / or splicing a second word into the second polypeptide, wherein the second word satisfies the second preset condition, wherein the second preset condition is that the contribution to toxicity is negative and the contribution to activity is positive; After adjusting the sequence of the second polypeptide based on the interpretability analysis results, the method further includes: The third polypeptide obtained after adjusting the second polypeptide is input into the polypeptide activity prediction model and the polypeptide toxicity prediction model again for verification to ensure that the adjusted third polypeptide is still predicted to be active and non-toxic; specifically, the adjusted third polypeptide is obtained; the activity information of the third polypeptide is obtained by using the polypeptide activity prediction model based on the sequence information and structural information of the third polypeptide, as well as the sequence information and structural information of the target receptor; and the toxicity information of the third polypeptide is predicted by using the polypeptide toxicity prediction model based on the sequence information, structural information and composition information of the third polypeptide, as well as the sequence information, structural information and composition information of the target receptor.

2. The method according to claim 1, characterized in that The polypeptide generation model includes one or more Transformer decoders, layer normalization, linear layers and softmax layers connected in sequence; The Transformer decoder includes a multi-head self-attention layer, a residual connection, a layer normalization, and a feedforward network layer.

3. The method according to claim 1, characterized in that The polypeptide activity prediction model includes a first Star-Transformer model, two first visual Transformer models, a first Transformer encoder, and multiple multi-layer perceptrons; The first Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor; The two first visual Transformer models are used to respectively extract features from the structural information of the first polypeptide and the structural information of the target receptor; the feature information respectively extracted by the two first visual Transformer models is respectively fed into the first Transformer encoder through a multi-layer perceptron for feature fusion; The outputs of the first Star-Transformer model and the first Transformer encoder are respectively fused after passing through a multi-layer perceptron, and the fused features enter the classifier constructed based on the multi-layer perceptron for prediction to obtain the activity information of the first polypeptide.

4. The method according to claim 3, characterized in that The input of the first visual Transformer model for extracting features from the structural information of the target receptor is a three-dimensional structural image block of the target receptor, and the input of the first visual Transformer model for extracting features from the structural information of the first polypeptide is a three-dimensional structural image block of the first polypeptide; The size of the target receptor three-dimensional structure image block is larger than the size of the first polypeptide three-dimensional structure image block.

5. The method according to claim 1, wherein The peptide toxicity prediction model includes a second Star-Transformer model, two second visual Transformer models, a third visual Transformer model, a second Transformer encoder, and multiple multi-layer perceptrons; The second Star-Transformer model is used to extract features from the sequence information of the first polypeptide and the sequence information of the target receptor; The two second visual Transformer models are used to respectively extract features from the structural information of the first polypeptide and the structural information of the target receptor; the feature information extracted by the two second visual Transformer models respectively passes through a multi-layer perceptron and enters the second Transformer encoder for feature fusion; The third visual Transformer model is used to extract features from the composition information of the first polypeptide and the composition information of the target receptor; The outputs of the second Star-Transformer model, the second Transformer encoder, and the third visual Transformer model are respectively fused after passing through a multi-layer perceptron. The fused features enter the classifier built based on the multi-layer perceptron for prediction to obtain the toxicity information of the first polypeptide.

6. The method according to claim 1, wherein After generating the first polypeptide based on the polypeptide generation model, the method further includes: Using an artificial intelligence model to predict the sequence information of the first polypeptide to obtain a structure prediction result; Based on the structure prediction result, a structural image of the first polypeptide is obtained as the structural information of the first polypeptide.

7. A polypeptide production device, characterized in that: The device comprises: A generation module, configured to generate a first polypeptide based on a polypeptide generation model; an activity prediction module, configured to obtain activity information of the first polypeptide based on the sequence information and structural information of the first polypeptide and the sequence information and structural information of the target receptor using a polypeptide activity prediction model; a toxicity prediction module, configured to predict toxicity information of the first polypeptide based on the sequence information, structural information, and composition information of the first polypeptide and the sequence information, structural information, and composition information of the target receptor using a polypeptide toxicity prediction model; An explanation module, used to perform interpretability analysis on the polypeptide activity prediction model and the polypeptide toxicity prediction model respectively; An adjustment module is configured to adjust the sequence of a second polypeptide based on the interpretability analysis result; the second polypeptide is the first polypeptide predicted to be active by the polypeptide activity prediction model and predicted to be non-toxic by the polypeptide toxicity prediction model; the adjustment module specifically comprises: a deleting unit, configured to delete a first word satisfying a first preset condition from the second polypeptide, wherein the first preset condition is that the contribution to toxicity is a positive value and the contribution to activity is a negative value; and / or a replacing unit, configured to replace the first word in the second polypeptide with a second word satisfying a second preset condition, wherein the second preset condition is that the contribution to toxicity is a negative value and the contribution to activity is a positive value; and / or a splicing unit, configured to splice a second word into the second polypeptide, wherein the second word satisfies the second preset condition, wherein the second preset condition is that the contribution to toxicity is a negative value and the contribution to activity is a positive value; The polypeptide generating device also includes: A module for re-inputting the third polypeptide obtained after adjusting the second polypeptide into the polypeptide activity prediction model and the polypeptide toxicity prediction model for verification to ensure that the third polypeptide obtained after adjustment is still predicted to be active and non-toxic; specifically comprising: an adjustment polypeptide acquisition module, for obtaining the adjusted third polypeptide; an activity verification module, for using the polypeptide activity prediction model to obtain the activity information of the third polypeptide based on the sequence information and structural information of the third polypeptide, as well as the sequence information and structural information of the target receptor; a toxicity verification module, for using the polypeptide toxicity prediction model to predict the toxicity information of the third polypeptide based on the sequence information, structural information and composition information of the third polypeptide, as well as the sequence information, structural information and composition information of the target receptor.

8. 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 polypeptide production method according to any one of claims 1 to 6 by executing the computer instructions.

Images