Prediction method for interaction between antibody and antigen and related equipment

Abstract

The embodiment of the invention provides a method for predicting interaction between an antibody and an antigen and related equipment, and belongs to the technical field of data processing. The method comprises the following steps: acquiring antibody sequence information of a target antibody and antigen sequence information of a target antigen; extracting antibody sequence features in the antibody sequence information, and extracting antigen sequence features in the antigen sequence information; determining respective bidirectional attention weights between the antibody sequence features and the antigen sequence features, and extracting bidirectional attention features between the antibody sequence features and the antigen sequence features based on the bidirectional attention weights; mapping the antibody sequence features to a spatial structure of a target antibody to obtain antibody structural features; and fusing the bidirectional attention features and the antibody structure features to obtain target features, and determining a target prediction result of interaction between the target antibody and the target antigen based on the target features. The prediction accuracy of the interaction between the target antibody and the target antigen can be improved.

Description

Technical Field

[0001] The present application relates to the field of data processing technology, and in particular to a method for predicting the interaction between an antibody and an antigen and related equipment. Background Art

[0002] The interaction between antibodies and antigens refers to the process in which antibodies (immunoglobulins produced by B lymphocytes or plasma cells) specifically recognize and bind to antigens (substances that can cause immune responses in the body, such as bacteria, viruses, toxins or foreign bodies). In order to provide a basis for medical research and other aspects, it is necessary to predict the interaction between the target antibody and the target antigen.

[0003] In the related technology, the interaction between the target antibody and the target antigen is predicted by inputting their three-dimensional spatial structure information. However, the three-dimensional spatial structure information is relatively complex, and such a prediction method requires a large amount of computing resources as support. As a result, when computing resources are limited, the related technology is often unable to conduct a comprehensive and in-depth analysis of the three-dimensional spatial structure information, which results in a low prediction accuracy of the interaction between the target antibody and the target antigen. Summary of the invention

[0004] The main purpose of the embodiments of the present application is to propose a method for predicting the interaction between an antibody and an antigen and related equipment, which can improve the prediction accuracy of the interaction between a target antibody and a target antigen when computing resources are limited.

[0005] To achieve the above objectives, the first aspect of the embodiments of the present application proposes a method for predicting the interaction between an antibody and an antigen, the method comprising:

[0006] Obtaining antibody sequence information of the target antibody and antigen sequence information of the target antigen;

[0007] Extracting antibody sequence features from antibody sequence information, and extracting antigen sequence features from antigen sequence information;

[0008] Determine respective bidirectional attention weights between the antibody sequence features and the antigen sequence features, and extract bidirectional attention features between the antibody sequence features and the antigen sequence features based on the respective bidirectional attention weights;

[0009] Mapping the antibody sequence features to the spatial structure of the target antibody to obtain the antibody structural features;

[0010] The bidirectional attention features and antibody structure features are fused to obtain the target features, and the target prediction results of the interaction between the target antibody and the target antigen are determined based on the target features.

[0011] In some embodiments, determining respective bidirectional attention weights between antibody sequence features and antigen sequence features, and extracting bidirectional attention features between the antibody sequence features and the antigen sequence features based on the respective bidirectional attention weights, comprises:

[0012] According to the preset first group of parameters, respectively determine the first query information, the first key information and the first value information corresponding to the antigen sequence information;

[0013] According to the preset second group parameters, second query information and second key information corresponding to the antibody sequence information are respectively determined;

[0014] Determining a first bidirectional attention weight characterizing the degree of attention of the target antigen to the target antibody based on the first query information and the second key information;

[0015] Determining a second bidirectional attention weight characterizing the degree of attention of the target antibody to the target antigen based on the second query information and the first key information;

[0016] The first value information is normalized according to the superimposed first bidirectional attention weight and the second bidirectional attention weight to obtain a bidirectional attention feature.

[0017] In some embodiments, determining a target prediction result of an interaction between a target antibody and a target antigen based on a target feature comprises:

[0018] The target prediction results are used to characterize at least one of a half-inhibitory concentration, an interaction class attribute, and a binding affinity index.

[0019] In some embodiments, the antibody sequence information is obtained by encoding and processing by an antibody sequence encoder, and the antigen sequence information is obtained by encoding and processing by an antigen sequence encoder;

[0020] Before obtaining the antibody sequence information of the target antibody and the antigen sequence information of the target antigen, it also includes:

[0021] Obtaining a first protein language model and a second protein language model trained by universal protein data, and obtaining sample antibody data and sample antigen data;

[0022] Training the first protein language model based on the sample antibody data until a preset first training condition is reached to obtain an antibody sequence encoder;

[0023] The second protein language model is trained based on the sample antigen data until a preset second training condition is reached to obtain an antigen sequence encoder.

[0024] In some embodiments, the antibody structural features are mapped from an antibody structure encoder;

[0025] The antibody structure encoder is trained through the following steps:

[0026] Obtaining sample sequence information and sample structure information corresponding to the sample antibody;

[0027] Input the sample sequence information into the antibody structure encoder, and input the sample structure information into the preset teacher network model;

[0028] The sample sequence information is mapped through the antibody structure encoder to obtain the student structure feature, and the sample structure information is mapped through the teacher network model to obtain the teacher structure feature;

[0029] A feature loss value is calculated based on the teacher structure features and the student structure features, and the antibody structure encoder is trained based on the feature loss value to obtain a trained antibody structure encoder.

[0030] In some embodiments, the feature loss value is calculated based on the teacher structural feature and the student structural feature, including:

[0031] Calculate the cosine similarity between the teacher's structural features and the student's structural features;

[0032] The exponential value corresponding to the cosine similarity is calculated based on the preset training parameters, and the logarithmic operation result of the exponential value is determined as the feature loss value.

[0033] In some embodiments, the target prediction result is predicted by a prediction model, and the prediction model includes a trained antibody sequence encoder, a trained antigen sequence encoder, and a trained antibody structure encoder;

[0034] The prediction model is trained through the following steps:

[0035] Obtaining first sequence information of sample antibodies and second sequence information of sample antigens, and verifying the prediction results;

[0036] Extracting a first sequence feature from the first sequence information, and extracting a second sequence feature from the second sequence information;

[0037] Determine respective sample bidirectional attention weights between the first sequence feature and the second sequence feature, and extract the sample bidirectional attention features between the first sequence feature and the second sequence feature based on the respective sample bidirectional attention weights;

[0038] Mapping the first sequence feature to the spatial structure of the sample antibody to obtain a first structural feature;

[0039] The sample bidirectional attention feature and the first structural feature are integrated to obtain the sample feature, and the sample prediction result of the interaction between the sample antibody and the sample antigen is determined based on the sample feature;

[0040] The model loss value is calculated based on the verification prediction results and the sample prediction results, the prediction model is trained according to the model loss value, and the trained prediction model is obtained.

[0041] In some embodiments, the model loss value is calculated based on the verification prediction result and the sample prediction result, including:

[0042] Calculating a first difference between a preset unit result and a verification prediction result, and calculating a second difference between a preset unit result and a sample prediction result;

[0043] Calculate a first logarithmic operation value of the sample prediction result, and determine the product of the verification prediction result and the first logarithmic operation value as a first multiplication value;

[0044] Calculate a second logarithmic operation value of the second difference, and determine the product of the first difference and the second logarithmic operation value as a second product value;

[0045] Based on the first multiplication value and the second multiplication value, a model loss value is obtained.

[0046] To achieve the above-mentioned purpose, a second aspect of the embodiments of the present application provides a device for predicting the interaction between an antibody and an antigen, the device comprising:

[0047] An acquisition module, used to acquire the antibody sequence information of the target antibody and the antigen sequence information of the target antigen;

[0048] A feature extraction module, used to extract antibody sequence features from antibody sequence information and to extract antigen sequence features from antigen sequence information;

[0049] A two-way attention feature calculation module, used to determine the respective two-way attention weights between the antibody sequence features and the antigen sequence features, and extract the two-way attention features between the antibody sequence features and the antigen sequence features based on the respective two-way attention weights;

[0050] A structural feature mapping module is used to map the antibody sequence features to the spatial structure of the target antibody to obtain the antibody structural features;

[0051] The target prediction module is used to fuse the bidirectional attention features and the antibody structure features to obtain the target features, and determine the target prediction results of the interaction between the target antibody and the target antigen based on the target features.

[0052] To achieve the above-mentioned purpose, the third aspect of an embodiment of the present application proposes an electronic device, which includes a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, the method for predicting the interaction between antibodies and antigens of the first aspect mentioned above is implemented.

[0053] To achieve the above objectives, the fourth aspect of the embodiments of the present application proposes a computer-readable storage medium, which stores a computer program. When the computer program is executed by a processor, it implements the method for predicting the interaction between antibodies and antigens according to the first aspect.

[0054] The method for predicting the interaction between an antibody and an antigen proposed in the present application and the related device, which obtains the antibody sequence information of the target antibody and the antigen sequence information of the target antigen; extracts the antibody sequence features in the antibody sequence information, and extracts the antigen sequence features in the antigen sequence information; determines the respective two-way attention weights between the antibody sequence features and the antigen sequence features, and extracts the two-way attention features between the antibody sequence features and the antigen sequence features based on the respective two-way attention weights; helps to more comprehensively understand the interaction between the target antibody and the target antigen by simultaneously considering the two-way attention features of the attention from the target antibody to the target antigen, and the attention from the target antigen to the target antibody; then, the antibody The antibody structure characteristics are mapped onto the spatial structure of the target antibody to obtain the antibody structure characteristics; the antibody structure characteristics can better reflect the functional characteristics of the target antibody, so that the prediction device can better analyze the interaction between the target antibody and the target antigen; then, the two-way attention characteristics and the antibody structure characteristics are fused to obtain the target characteristics, and the target prediction result of the interaction between the target antibody and the target antigen is determined based on the target characteristics; without inputting the spatial structure information related to the target antibody into the prediction device, the prediction device proposed in the present application can process the input antibody sequence information and antigen sequence information based on limited computing resources, thereby improving the prediction accuracy of the interaction between the target antibody and the target antigen. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] Figure 1 This is a schematic diagram of an optional application scenario of the device for predicting the interaction between antibodies and antigens provided in an embodiment of the present application;

[0056] Figure 2 is an optional flow chart of the method for predicting the interaction between an antibody and an antigen provided in the embodiments of the present application;

[0057] Figure 3 yes Figure 2 A flowchart of an implementation of step 103 in FIG.

[0058] Figure 4This is a schematic diagram of an optional two-way attention module processing flow of the method for predicting the interaction between an antibody and an antigen provided in an embodiment of the present application;

[0059] Figure 5 yes Figure 2 An implementation flow chart before step 101 in FIG.

[0060] Figure 6 This is an optional flow chart of antibody structure encoder training steps for the method for predicting the interaction between an antibody and an antigen provided in an embodiment of the present application;

[0061] Figure 7 This is an optional antibody structure encoder training schematic diagram of the method for predicting the interaction between an antibody and an antigen provided in an embodiment of the present application;

[0062] Figure 8 yes Figure 6 A flowchart of an implementation of step 504 in FIG.

[0063] Fig. 9 It is a flow chart of the training steps of an optional prediction model of the method for predicting the interaction between an antibody and an antigen provided in the embodiment of the present application;

[0064] Fig.10 yes Fig. 9 A flowchart of an implementation of step 706 in FIG.

[0065] Fig.11 This is an optional flow chart of the device for predicting the interaction between an antibody and an antigen provided in the embodiments of the present application;

[0066] Fig.12 It is a schematic diagram of the hardware structure of the electronic device provided in the embodiment of the present application. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0068] It should be noted that, although the functional modules are divided in the device schematic diagram and the logical order is shown in the flowchart, in some cases, the steps shown or described may be performed in a different order than the module division in the device or the order in the flowchart. The terms "first", "second", etc. in the specification, claims and the above drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence.

[0069] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art to which this application belongs. The terms used herein are only for the purpose of describing the embodiments of this application and are not intended to limit this application.

[0070] First, some nouns involved in this application are analyzed:

[0071] B lymphocytes (also called B cells) are derived from pluripotent stem cells in the bone marrow. The progenitor cells of B lymphocytes are present in the hematopoietic cell islands of the fetal liver (14 days in embryonic mice or 8-9 weeks in infants), after which the production and differentiation sites of B lymphocytes are gradually replaced by the bone marrow. Mature B cells mainly reside in the lymph nodules in the superficial layer of the lymph node cortex and the lymph nodules in the red and white pulp of the spleen. B cells can differentiate into plasma cells under antigen stimulation. Plasma cells can synthesize and secrete antibodies (immunoglobulins), and mainly perform the body's humoral immunity.

[0072] Ribonucleic acid (RNA) is a genetic information carrier present in biological cells and some viruses and viroids; the main function of RNA in the body is to guide protein synthesis. RNA is a long chain molecule formed by ribonucleotides condensed by phosphodiester bonds, in which a ribonucleotide molecule is composed of phosphate, ribose and bases. There are four main bases in RNA, namely adenine (A), guanine (G), cytosine (C) and uracil (U), among which U replaces T (thymine) in DNA.

[0073] Deoxyribonucleic acid (DNA) is one of the four biological macromolecules contained in biological cells. DNA carries the genetic information necessary for the synthesis of RNA and proteins and is an essential biological macromolecule for the development and normal functioning of organisms.

[0074] Complementary deoxyribonucleic acid (cDNA) is a DNA chain that is complementary to RNA after reverse transcription in vitro. Unlike genomic DNA (gDNA), cDNA has no introns but only exon sequences.

[0075] Natural language processing (NLP): NLP uses computers to process, understand and apply human languages ​​(such as Chinese and English). NLP is a branch of artificial intelligence and an interdisciplinary subject between computer science and linguistics. It is often referred to as computational linguistics. Natural language processing includes grammatical analysis, semantic analysis, and text understanding. Natural language processing is often used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining. It involves data mining related to language processing, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research, and linguistic research related to language computing.

[0076] The interaction between antibodies and antigens refers to the process in which antibodies (immunoglobulins produced by B lymphocytes or plasma cells) specifically recognize and bind to antigens (substances that can cause immune responses in the body, such as bacteria, viruses, toxins or foreign bodies). In order to provide a basis for medical research and other aspects, it is necessary to predict the interaction between the target antibody and the target antigen.

[0077] In the related technology, the interaction between the target antibody and the target antigen is predicted by inputting their three-dimensional spatial structure information. However, the three-dimensional spatial structure information is relatively complex, and such a prediction method requires a large amount of computing resources as support. As a result, when computing resources are limited, the related technology is often unable to conduct a comprehensive and in-depth analysis of the three-dimensional spatial structure information, which results in a low prediction accuracy of the interaction between the target antibody and the target antigen.

[0078] Based on this, the embodiments of the present application provide a method for predicting the interaction between an antibody and an antigen and related equipment, which can improve the prediction accuracy of the interaction between a target antibody and a target antigen when computing resources are limited.

[0079] It should be noted that in the embodiments of the present application, when it comes to information related to user characteristics such as user basic information or user identity, the user's permission or consent will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the user's separate permission or separate consent will be obtained first. After the separate permission or separate consent of the user is clearly obtained, the necessary data for enabling the normal operation of the embodiments of the present application will be obtained. For example, before analyzing the antibody sequence information and antigen sequence information related to the user, the embodiments of the present application will first obtain the consent of the relevant user; or, before obtaining the first sequence information and the second sequence information related to the user for training, the embodiments of the present application will also obtain the consent of the relevant user, otherwise the antibody sequence information, antigen sequence information, first sequence information and second sequence information used in the embodiments of the present application cannot be obtained.

[0080] The method for predicting the interaction between an antibody and an antigen and the related device provided in the embodiments of the present application are specifically described by the following embodiments, wherein the related device includes a prediction device for the interaction between an antibody and an antigen (hereinafter referred to as "prediction device" for ease of description). First, the application scenario of the prediction device in the embodiments of the present application is described, such as Figure 1 As shown, Figure 1 This is an optional application scenario diagram of the prediction device for the interaction between antibodies and antigens provided in the embodiment of the present application. First, the acquired antibody sequence information and antigen sequence information are input into the prediction device in the embodiment of the present application; the prediction device can extract the antibody sequence features and antigen sequence features respectively based on the antibody sequence information and the antigen sequence information, and obtain the two-way attention features by capturing the dependency between the antibody sequence features and the antigen sequence features; then, the prediction device can also map the antibody sequence features to the spatial structure of the target antibody, thereby obtaining the antibody structural features; then, the target features are obtained by fusing the two-way attention features and the antibody structural features; finally, the interaction between the target antibody and the target antigen is predicted based on the target features to obtain the target prediction result. It can be understood that without inputting the spatial structure information related to the target antibody into the prediction device, the prediction device can process the input antibody sequence information and antigen sequence information based on limited computing resources, and obtain the target prediction result with higher prediction accuracy.

[0081] After understanding the application scenarios of the prediction device proposed in the present application, the prediction method for the interaction between antibodies and antigens proposed in the embodiments of the present application (for the sake of convenience, it may also be referred to as "prediction method" below) is described.

[0082] In the embodiments of the present application, the prediction device will be described from the perspective of the prediction device, which can be integrated into a computer device, such as a server. Figure 2 As shown, Figure 2 is an optional flow chart of the method for predicting the interaction between an antibody and an antigen provided in the embodiments of the present application, Figure 2 The method in the embodiment may include but is not limited to the following steps 101 to 105. When the prediction device executes the prediction method, the specific process is as follows. It should be noted that the present embodiment Figure 2 The order of step 101 to step 105 is not specifically limited, and the order of steps can be adjusted or some steps can be reduced or added according to actual needs.

[0083] Step 101, obtaining antibody sequence information of a target antibody and antigen sequence information of a target antigen.

[0084] The following is a detailed description of step 101.

[0085] Among them, antibody (Anti body) is a protein produced by B lymphocytes, which can specifically recognize and bind to foreign substances (such as bacteria, viruses or other pathogens) or abnormal cells in the body (such as cancer cells). This binding can trigger a series of immune responses, including neutralizing pathogens, promoting pathogens to be cleared by phagocytes or activating the complement system. Target antibody refers to the antibody used in the embodiments of this application.

[0086] Among them, antibody sequence information refers to the amino acid sequence that constitutes the antibody protein, which determines the specificity and function of the antibody; antibody sequence information reveals in detail the arrangement order of amino acids in the antibody protein, and thus can play a key role in predicting the spatial conformation of the antibody and antibody-related functions.

[0087] Furthermore, antibody sequence information can be obtained from patients, experimental animals or cell cultures. These substances are the source of antibody generation, and the DNA or RNA inside them contains antibody sequence information; alternatively, antibody sequence information can be obtained by polymerase chain reaction technology, amplifying antibody sequence fragments in cDNA and sequencing them; of course, antibody sequence information can also be obtained by a specific method selected according to actual conditions, and the embodiments of the present application do not limit this.

[0088] Among them, antigen refers to a substance that can stimulate the body to produce an immune response, usually some components of foreign substances (such as pathogens) or molecules on the surface of abnormal cells in the body. Antigen has immunogenicity and reactogenicity, the former refers to the ability of antigens to stimulate the body to produce an immune response, and the latter refers to the ability of antigens to specifically bind to immune response products (such as antibodies). Target antigen refers to the antigen used in the examples of this application.

[0089] Among them, antigen sequence information refers to the sequence of molecules that constitute antigens (such as proteins, polysaccharides, nucleic acids, etc.). This sequence determines the specificity and immunogenicity of the antigen; antigen sequence information reveals in detail the arrangement order of the units that constitute its structure in the antigen molecule. For example, for protein antigens, the antigen sequence information is the amino acid sequence; the antigen sequence information of polysaccharides is the monosaccharide sequence; the antigen sequence information of nucleic acid antigens is the nucleotide sequence. Antigen sequence information can play a key role in predicting the spatial conformation of antigens and antigen-related functions.

[0090] Furthermore, antigen sequence information can be obtained by sequencing proteins, polypeptides or other types of molecules; or, it can be obtained by obtaining its encoding gene through gene cloning technology and sequencing the encoding gene; of course, antigen sequence information can also be obtained by a specific method selected according to actual conditions, and the embodiments of the present application do not limit this.

[0091] Furthermore, the method for sequencing the target antibody and target antigen can be proteomics technology, mass spectrometry, peptide mapping analysis, etc. The specific determination method can be selected according to the actual situation, and the embodiment of the present application does not limit the selection of the sequencing method.

[0092] Step 102: extracting antibody sequence features from the antibody sequence information, and extracting antigen sequence features from the antigen sequence information.

[0093] The following is a detailed description of step 102.

[0094] In some embodiments, an antibody sequence encoder can be used to extract antibody sequence features from antibody sequence information; wherein the antibody sequence encoder is a tool or model for processing and parsing antibody sequence information, which can convert antibody sequences into a computer-understandable format and be used for various downstream target tasks.

[0095] Among them, the antibody sequence encoder processes the antibody sequence through its internal mechanism (such as convolution operation, recurrent unit, etc.), automatically learns and extracts the key features in the sequence, and the key features include local features and global features of the antibody sequence information; then, after completing the processing of the antibody sequence information, the antibody sequence encoder outputs the encoding information containing one or more feature vectors (i.e., antibody sequence features).

[0096] Furthermore, the antibody sequence encoder may be an encoder that relies on predefined rules and algorithms to parse the antibody sequence, which typically uses a standard numbering system to annotate key regions in the antibody sequence information, including complementary determining regions (CDRs) and framework regions (FRs); alternatively, a machine learning encoder trained with a general algorithm may be used, for example, the antibody sequence encoder may be an antibody-specific bidirectional encoder based on transformation (Antibody-specific Bidirectional Encoder Representations from Transformers, AntiBERTa); of course, the antibody sequence encoder may also be specifically selected according to actual conditions, and the embodiments of the present application do not limit this.

[0097] Furthermore, before the antibody sequence information is input into the antibody sequence encoder, the antibody sequence information may be preprocessed, including but not limited to removing irrelevant characters (such as non-standard amino acid codes), unifying the sequence format, padding or truncating to a uniform length (if the antibody sequence encoder requires a fixed length input), etc.; the specific preprocessing method may also be selected according to actual conditions, and the embodiments of the present application are not limited to this.

[0098] Step 103, determining respective bidirectional attention weights between the antibody sequence features and the antigen sequence features, and extracting bidirectional attention features between the antibody sequence features and the antigen sequence features based on the respective bidirectional attention weights.

[0099] The following is a detailed description of step 103.

[0100] In some embodiments, by simultaneously considering the attention from the target antibody to the target antigen, and the attention from the target antigen to the target antibody, this bidirectional attention can help the prediction device to more comprehensively understand the interaction between the target antibody and the target antigen, thereby obtaining a bidirectional attention feature. That is, the bidirectional attention feature can capture the characteristics between the antibody sequence information and the antigen sequence information that change over time or during the binding process, so as to determine the target feature based on the bidirectional attention feature, thereby helping to improve the prediction accuracy of the interaction between the target antibody and the target antigen.

[0101] Among them, the bidirectional attention feature is calculated based on the bidirectional attention weights corresponding to the target antibody and the target antigen. The bidirectional attention weight refers to the attention value assigned by the prediction device to each element (such as amino acid residues) in the two sequences when considering the antibody sequence information and the antigen sequence information at the same time. Usually, the bidirectional attention weight changes dynamically with the difference of the antibody sequence information and the antigen sequence information.

[0102] In some embodiments, Figure 3 As shown, Figure 3 yes Figure 2 A flowchart for implementing step 103 in the method of FIG. 1 , wherein the respective bidirectional attention weights between the antibody sequence features and the antigen sequence features are determined, and the bidirectional attention features between the antibody sequence features and the antigen sequence features are extracted based on the respective bidirectional attention weights, including the following steps 201 to 205:

[0103] Step 201, according to a preset first group of parameters, first query information, first key information and first value information corresponding to the antigen sequence information are respectively determined.

[0104] Step 202: Determine the second query information and the second key information corresponding to the antibody sequence information according to the preset second group parameters.

[0105] Step 203: Determine a first bidirectional attention weight that characterizes the degree of attention of the target antigen to the target antibody based on the first query information and the second key information.

[0106] Step 204: Determine a second bidirectional attention weight that characterizes the degree to which the target antibody pays attention to the target antigen based on the second query information and the first key information.

[0107] Step 205, normalize the first value information according to the superimposed first bidirectional attention weight and the second bidirectional attention weight to obtain a bidirectional attention feature.

[0108] Steps 201 to 205 are described in detail below.

[0109] like Figure 4 As shown, Figure 4 It is a schematic diagram of an optional two-way attention module processing flow of the method for predicting the interaction between antibodies and antigens provided in an embodiment of the present application, wherein the prediction device may include a two-way attention module, and a two-way attention mechanism is provided in the two-way attention module. The two-way attention module is used to perform two-way attention processing on the input antibody sequence features and then the antigen sequence features, and output the two-way attention features.

[0110] It should be noted that Figure 4The letters in the antibody sequence information and antigen sequence information are used to represent specific amino acids, for example, G represents glycine (Gly), K represents lysine (Lysine), A represents alanine (Alanine), T represents threonine (Threonine), S represents serine (Serine), V represents valine (Val ine), Q represents glutamine (Glutamine), etc.; Figure 4 The antibody sequence information and antigen sequence information are only used as examples and can be set according to actual conditions. The embodiments of the present application do not limit this.

[0111] For example, Figure 4 As shown, the antibody sequence information Antigen sequence information input into the prediction device; then, the prediction device respectively and Perform feature extraction to obtain residue-level embedded antibody sequence features and antigen sequence characteristics Among them, l Ag and l Ab are the target antigen and target antibody and sequence length, respectively, s is the hidden layer output dimension of the prediction device composed of the neural network; then, Projected into the query vector (query), key (key) and value vector (value) space, the first query information, first key information and first value information corresponding to the antigen sequence information are obtained:

[0112]

[0113] in, For the first query information, For the first key information, is the first value information; and All are preset first group parameters. In addition, residue-level embedding refers to the process of taking the amino acid residues in the protein (amino acids that have lost some chemical groups due to the formation of peptide bonds in the polypeptide chain) as basic units and converting them into numerical representations or embedding vectors through some methods (such as machine learning, deep learning, etc.). The embedding vector can capture information such as the chemical properties, physical properties, and interactions with the surrounding environment of the residues.

[0114] At the same time, Projected into the query vector and key space, the second query information and second key information corresponding to the antibody sequence information are obtained:

[0115]

[0116] in, For the second query information, is the second key information; and These are the default second group parameters.

[0117] It should be noted that the specific values ​​of the first group of parameters and the second group of parameters can be set according to actual conditions, and the embodiments of the present application do not limit this.

[0118] Furthermore, the bidirectional attention feature context is determined according to the following formula (1):

[0119]

[0120] Where s is a scoring function, which is a pre-set function that evaluates the rationality of an object or pattern by calculating the score of a specific parameter or attribute; based on the scoring function s, the first bidirectional attention weight representing the degree of attention of the target antigen to the target antibody is calculated And the second two-way attention weight representing the degree of attention of the target antibody to the target antigen is calculated in, and In order to facilitate matrix operations and processing, the first bidirectional attention weight is transposed to obtain Then superimpose the first bidirectional attention weight and the second bidirectional attention weight to obtain

[0121] Furthermore, the first bidirectional attention weight and the second bidirectional attention weight are normalized using a normalized exponential function to obtain Multiply the normalized result and the first value information to obtain the bidirectional attention feature

[0122] Among them, the normalized exponential function (softmax function) is used to convert the output of the neural network into a probability distribution. Specifically, it can "compress" a K-dimensional vector z containing any real number into another K-dimensional real vector σ(z), so that the range of each element is between (0,1), and the sum of all elements is 1, and σ is the symbol of the normalization function.

[0123] It should be noted that, since the embodiment of the present application pays more attention to the degree of attention of antigen sequence information to antibody sequence information, the output bidirectional attention feature is strongly correlated with the first value information; in another embodiment, the output bidirectional attention feature can also be strongly correlated with the second value information, wherein the second value information is generated by the prediction device Projected onto the value vector; alternatively, the output bidirectional attention feature may include features that are strongly correlated with the first value information and features that are strongly correlated with the second value information; the output setting of the bidirectional attention feature may be specifically performed according to actual conditions, and the embodiments of the present application do not impose any restrictions on this.

[0124] Step 104, mapping the antibody sequence features to the spatial structure of the target antibody to obtain the antibody structural features.

[0125] Step 104 is described in detail below.

[0126] In some embodiments, since the sequence characteristics of the protein correspond to the structure of the target antibody to a certain extent, after obtaining the antibody sequence characteristics, the prediction device can map them to the spatial structure of the target antibody and obtain the antibody structural characteristics; the antibody structural characteristics can better reflect the functional characteristics of the target antibody, making it easier for the prediction device to better analyze the interaction between the target antibody and the target antigen.

[0127] Among them, antibody sequence features such as highly conserved regions, specific amino acid patterns, and secondary structure predictions all contain important information about the three-dimensional structure of proteins. For example, certain combinations of amino acid residues are highly conserved during evolution, which often means that they play a key role in maintaining protein structure or function. In addition, by comparing sequence features, we can find similarities between different proteins and infer that they may have similar structures or functions.

[0128] In another embodiment, in addition to obtaining the antibody structural features, the antigen sequence features can also be simultaneously mapped to the spatial structure of the target antigen to obtain the antigen structural features, so as to facilitate the subsequent fusion of the antigen structural features, antibody structural features and bidirectional attention features to obtain the target features, thereby further improving the accuracy of the obtained target features.

[0129] Step 105, the bidirectional attention feature and the antibody structure feature are integrated to obtain the target feature, and the target prediction result of the interaction between the target antibody and the target antigen is determined based on the target feature.

[0130] The following is a detailed description of step 105.

[0131] In some embodiments, after obtaining the bidirectional attention feature and the antibody structural feature, the prediction device can fuse the bidirectional attention feature and the antibody structural feature to obtain the target feature; it can be understood that the bidirectional attention feature dynamically focuses on the key sequence information of the interaction between the target antibody and the target antigen, and the antibody structural feature reflects the three-dimensional shape, surface properties and information related to its binding ability with the target antigen of the target antibody; in this way, the prediction device can make a high-accuracy prediction of the interaction between the target antibody and the target antigen based on the target feature fused with this information, and thus obtain the target prediction result.

[0132] In some embodiments, determining a target prediction result of an interaction between a target antibody and a target antigen based on a target feature comprises the following steps 301:

[0133] Step 301 : The target prediction result is used to characterize at least one of a half inhibition concentration, an interaction category attribute, and a binding affinity index.

[0134] The following is a detailed description of step 301.

[0135] Among them, the half-inhibitory concentration (also known as the half-inhibition rate, IC50) refers to the concentration of a substance required to cause a certain biological effect (such as enzyme activity, cell proliferation, etc.) to be inhibited by 50% under specific experimental conditions; in medical research, IC50 is often used to evaluate the inhibitory effect of drugs on target molecules and is an important indicator for measuring drug efficacy.

[0136] Among them, the interaction category attribute refers to the characteristics that describe the type of interaction between two or more molecules (such as antibodies and antigens, drugs and targets, etc.). These interactions can include physical binding, chemical reactions, signal transduction and other types.

[0137] Among them, the binding affinity index is an indicator to measure the binding strength between two molecules. In the interaction between antibody and antigen, the binding affinity index reflects the antibody's ability to recognize and bind to the antigen.

[0138] It should be noted that the target prediction result can be set in a specific form of expression according to the actual situation. For example, the target prediction result can be displayed in the form of numerical value, chart or classification label, etc., and the embodiment of the present application does not limit this. In addition, according to the different target tasks of determining the interaction between the target antibody and the target antigen, the meaning represented by the target prediction result can also be adjusted accordingly; or, when there are multiple target tasks, the target prediction result can be displayed as a composite result, for example, the target task includes a classification task and a regression task, and the target prediction result is first pooled along the length direction of the antibody; after that, the pooled results are respectively input into the output fully connected layer for the regression task and the classification task to predict the regression value and classification result of a specific property, and are shown by the interaction category attribute and the binding affinity index.

[0139] In some embodiments, Figure 5 As shown, Figure 5 yes Figure 2 A flowchart for implementing the method before step 101 in the embodiment of the present invention, wherein before obtaining the antibody sequence information of the target antibody and the antigen sequence information of the target antigen, the following steps 401 to 403 are also included:

[0140] Step 401 : Obtain a first protein language model and a second protein language model trained by general protein data, and obtain sample antibody data and sample antigen data.

[0141] Step 402 : training a first protein language model based on sample antibody data until a preset first training condition is reached to obtain an antibody sequence encoder.

[0142] Step 403: Train the second protein language model based on the sample antigen data until a preset second training condition is reached to obtain an antigen sequence encoder.

[0143] Steps 401 to 403 are described in detail below.

[0144] In some embodiments, the prediction device may include an antibody sequence encoder and an antigen sequence encoder, and the antibody sequence information is obtained by encoding and processing the antibody sequence encoder, and the antigen sequence information is obtained by encoding and processing the antigen sequence encoder. Before the prediction device is put into practical application, the antibody sequence encoder and the antigen sequence encoder can be trained first, and then the antibody sequence encoder used in the embodiment of the present application for feature extraction of the antibody sequence information and the antigen sequence encoder used for feature extraction of the antigen sequence information can be obtained.

[0145] Furthermore, both the first protein language model and the second protein language model are models that use natural language processing (NLP) technology to analyze and understand protein sequences. This type of model regards protein sequences as "sentences" composed of amino acids, drawing on the successful experience of language models in text processing, thereby realizing functions such as prediction, generation, functional domain identification, and structural understanding of protein sequences. In addition, both the first protein language model and the second protein language model are trained by general protein data. General protein data refers to various types of information and data related to proteins. These data can come from biological experiments, computational simulations, databases, and other aspects. General protein data covers multiple levels such as protein structure, function, interaction, and evolutionary relationship.

[0146] Furthermore, since the antibody sequence encoder in the embodiment of the present application focuses on extracting antibody sequence features from antibody sequence information, and the antigen sequence encoder focuses on extracting antigen sequence features from antigen sequence information, the antibody sequence encoder and the antigen sequence encoder can be respectively trained for features using the acquired sample antibody data and sample antigen data.

[0147] Furthermore, the sample antibody data is a set of data sets containing one or more antibody sequences, which can be collected from biological experiments, bioinformatics databases or other sources; similarly, the sample antigen data is a set of data sets containing one or more antigen sequences, which can also be collected from biological experiments, bioinformatics databases and other sources. It should be noted that the embodiments of the present application do not limit the legal sources of the sample antibody data and sample antigen data, and the number of the collected sample antibody data and sample antigen data.

[0148] Furthermore, the antibody sequence encoder is trained based on the acquired sample antibody data, and the relevant parameters of the antibody sequence encoder are adjusted according to the first training result obtained, until the preset first training condition is reached, and a trained antibody sequence encoder is obtained; similarly, the antibody sequence encoder is trained based on the acquired sample antigen data, and the relevant parameters of the antigen sequence encoder are adjusted according to the second training result obtained, until the preset second training condition is reached, and a trained antigen sequence encoder is obtained; the relevant parameters of the first sequence encoder and the second sequence encoder may include but are not limited to the number of encoder layers, hidden unit size, learning rate, batch size, etc., which can be set according to actual conditions, and the embodiments of the present application are not limited to this.

[0149] Further, the first training condition and the second training condition (collectively referred to as "training condition") can be set according to actual conditions, and the training condition can be that the number of training iterations reaches a preset number threshold, or the training time reaches a preset time threshold, etc., and the embodiment of the present application does not limit this. In addition, the trained antibody sequence encoder and the trained antigen sequence encoder can be integrated in the prediction device, so that the prediction device uses the trained antibody sequence encoder to perform feature extraction operations on the input antibody sequence information to obtain antibody sequence features; and uses the trained antigen sequence encoder to perform feature extraction operations on the input antigen sequence information to obtain antigen sequence features.

[0150] In some embodiments, Figure 6 As shown, Figure 6 This is an optional flow chart of antibody structure encoder training steps for the method for predicting the interaction between an antibody and an antigen provided in an embodiment of the present application. The antibody structure encoder is trained by following steps 501 to 504:

[0151] Step 501: Obtain sample sequence information and sample structure information corresponding to the sample antibody.

[0152] Step 502, input the sample sequence information into the antibody structure encoder, and input the sample structure information into the preset teacher network model.

[0153] Step 503, mapping the sample sequence information through the antibody structure encoder to obtain the student structure feature, and mapping the sample structure information through the teacher network model to obtain the teacher structure feature.

[0154] Step 504, a feature loss value is calculated based on the teacher structure feature and the student structure feature, and an antibody structure encoder is trained based on the feature loss value to obtain a trained antibody structure encoder.

[0155] Steps 501 to 504 are described in detail below.

[0156] In some embodiments, Figure 7 As shown, Figure 7 It is an optional antibody structure encoder training diagram of the method for predicting the interaction between antibodies and antigens provided in an embodiment of the present application. Specifically, the sample sequence information obtained based on the sample antibody correspondence is input into the antibody structure encoder (student network model), and the sample structure information obtained based on the sample antibody correspondence is input into the teacher network model, and the antibody structure encoder is distilled and trained by the teacher network model.

[0157] The sample antibody and the target antibody are usually antibodies with similar functions and structures, and the sample antibody can be obtained through legal databases or legal experiments. Then, feature extraction is performed based on the pre-trained antibody sequence encoder to obtain sample sequence information; and feature extraction is performed based on the pre-trained antigen sequence encoder to obtain sample antigen information.

[0158] Furthermore, the teacher network model is a pre-trained model that has the ability to extract three-dimensional structural information from sample antibodies (including simple antibodies and complex antibodies). Furthermore, the features extracted by the teacher network model will serve as "teacher" knowledge to guide the learning of the student network model (i.e., antibody structure encoder). In addition, the teacher network model can be obtained by training with general protein data or based on specific sample antibody data. The embodiment of the present application does not limit the selection of the teacher network model.

[0159] Furthermore, the student structural feature is the feature output by the antibody structural encoder after processing the sample sequence information. The student structural feature attempts to capture the structural information of the sample antibody, but is not as rich and accurate as the feature directly extracted from the three-dimensional structural information. The teacher structural feature is the feature output by the teacher network model after processing the sample structural information. It represents the real three-dimensional structural information of the antibody. The teacher structural feature can be used as a "standard answer" to train the antibody structural encoder to obtain an antibody structural encoder with high-accuracy mapping capabilities.

[0160] Furthermore, a feature loss value is calculated based on the teacher structural features and the student structural features obtained by mapping. The feature loss value can be calculated by the mean square error method, the Manhattan distance method, etc. The embodiment of the present application does not limit the specific calculation method of the feature loss value.

[0161] Furthermore, the teacher network model can be used only in the training process of the antibody structure encoder, and the trained antibody structure encoder can be integrated into the prediction device together with the trained antigen sequence encoder and antibody sequence encoder to map the antibody sequence features to the spatial structure of the target antibody in practical applications to obtain the antibody structural features.

[0162] In some embodiments, Figure 8 As shown, Figure 8 yes Figure 6 A flowchart for implementing step 504 in FIG. 1 , wherein the feature loss value is calculated based on the teacher structural feature and the student structural feature, including the following steps 601 to 602:

[0163] Step 601, calculating the cosine similarity between the teacher's structural features and the student's structural features.

[0164] Step 602, calculating the exponential value corresponding to the cosine similarity based on the preset training parameters, and determining the logarithmic operation result of the exponential value as the feature loss value.

[0165] Steps 601 to 602 are described in detail below.

[0166] In some embodiments, the characteristic loss value can be calculated by the following formula (2):

[0167]

[0168] Among them, i represents the i-th student structural feature corresponding to the sample antibody, j represents the j-th teacher structural feature corresponding to the sample antibody, and i and j are only used for distinction; τ is the preset training parameter; exp is the exponential function, and the value calculated based on the exponential function is called the exponential value; log is the logarithmic function, and the result calculated based on the logarithmic function is called the logarithmic operation result.

[0169] The cosine similarity between the teacher's structural features and the student's structural features is calculated by the following formula (3):

[0170]

[0171] Specifically, the antibody sequence encoder and the antibody structure encoder are denoted as f seq , let the teacher network encoder be f str , the sample antibody x i Corresponding sample sequence characteristics and sample structure characteristics Enter f respectively seq and f str In, f seq Output student structure characteristics f str Output teacher structure characteristics The training objective of the teacher network model is to maximize the similarity between representation pairs corresponding to the same antibody and minimize the similarity between unpaired representations.

[0172] In some embodiments, Fig. 9 As shown, Fig. 9 701 to 706:

[0173] Step 701, obtaining the first sequence information of the sample antibody and the second sequence information of the sample antigen, and verifying the prediction result.

[0174] Step 702: extract the first sequence feature in the first sequence information, and extract the second sequence feature in the second sequence information.

[0175] Step 703: determine the respective sample bidirectional attention weights between the first sequence feature and the second sequence feature, and extract the sample bidirectional attention features between the first sequence feature and the second sequence feature based on the respective sample bidirectional attention weights.

[0176] Step 704: Map the first sequence feature to the spatial structure of the sample antibody to obtain a first structural feature.

[0177] Step 705: The sample bidirectional attention feature and the first structural feature are integrated to obtain the sample feature, and a sample prediction result of the interaction between the sample antibody and the sample antigen is determined based on the sample feature.

[0178] Step 706, a model loss value is calculated based on the verification prediction result and the sample prediction result, a prediction model is trained according to the model loss value, and a trained prediction model is obtained.

[0179] Steps 701 to 706 are described in detail below.

[0180] In some embodiments, the target prediction result is obtained by prediction by a prediction model, and the prediction model may include a trained antibody sequence encoder, a trained antigen sequence encoder, and a trained antibody structure encoder. That is, by jointly training the antibody sequence encoder, the antigen sequence encoder, and the antibody structure encoder in the prediction model, a prediction model for outputting the target prediction result is obtained.

[0181] Further, the first sequence information of the sample antibody used to characterize the antibody sequence information and the second sequence information of the sample antigen used to characterize the antigen sequence information are obtained. The method for obtaining the first sequence information and the second sequence information is similar to step 101 and will not be repeated here. In addition, it is also necessary to obtain a verification prediction result, which is a known result of the sample antibody-sample antigen interaction verified by the experiment, and is used for the subsequent evaluation of the accuracy of the prediction model prediction.

[0182] Furthermore, steps 702 to 705 are similar to steps 102 to 105, except that the names are different. The "first" in steps 702 to 705 is used to characterize the content related to the antibody, and the "second" in steps 702 to 705 is used to characterize the content related to the antigen. In order to avoid redundant description, it will not be repeated here.

[0183] Furthermore, the sample prediction results are obtained by jointly processing the antibody sequence encoder, antigen sequence encoder and antibody structure encoder in the prediction model. Therefore, after obtaining the sample prediction results, the model loss value can be calculated based on the verification prediction results and the sample prediction results, and the various parts of the prediction model can be jointly adjusted based on the model loss value to obtain the trained prediction model.

[0184] Further, joint adjustment refers to the synchronous adjustment of the relevant parameters of the antibody sequence encoder, antigen sequence encoder and antibody structure encoder based on the model loss value after the model loss value is calculated. The prediction model obtained by the joint adjustment can show stronger stability and adaptability, that is, stronger robustness and stronger generalization ability. Among them, the joint adjustment includes adjusting the parameters of each encoder, that is, the parameters such as the regularization value, learning rate, number of iterations, encoder weights of each encoder can be adjusted synchronously, and the specific parameter values ​​can be adjusted according to the actual situation, and the embodiments of the present application are not limited to this.

[0185] In some embodiments, the prediction model and the module used to execute the downstream target tasks can be further jointly trained according to the different downstream target tasks connected to the prediction model; specifically, a joint loss value is calculated based on the output result of the downstream target task, and then the various parts of the prediction model are jointly adjusted according to the joint loss value to obtain a trained prediction model.

[0186] In some embodiments, Fig.10 As shown, Fig.10 yes Fig. 9 A flowchart for implementing step 706 in FIG. 1 , wherein the model loss value is calculated based on the verification prediction result and the sample prediction result, including the following steps 801 to 804:

[0187] Step 801, calculate the first difference between the preset unit result and the verification prediction result, and calculate the second difference between the preset unit result and the sample prediction result.

[0188] Step 802, calculate a first logarithmic operation value of the sample prediction result, and determine that the product of the verification prediction result and the first logarithmic operation value is a first product value.

[0189] Step 803, calculating a second logarithmic operation value of the second difference, and determining the product of the first difference and the second logarithmic operation value as a second product value.

[0190] Step 804, obtaining a model loss value based on the first multiplication value and the second multiplication value.

[0191] Steps 801 to 804 are described in detail below.

[0192] In some embodiments, the model loss value BCE can be calculated by the following formula (4):

[0193]

[0194] Among them, 1 is the preset unit result; To verify the prediction results; i Predict the results for the sample; is the first difference; (1-y i ) is the second difference; is the first logarithmic operation value; is the first multiplication value; log(1-y i ) is the second logarithmic operation value; (1-y i )log(1-y i ) is the second multiplication value; the verification prediction result and the sample prediction result are used to characterize the probability value that the corresponding sample antibody and sample antigen can interact with each other; N is used to represent the total number of verification / sample prediction results, and i is used to represent the i-th verification / sample prediction result.

[0195] like Fig.11 As shown, Fig.11 This is an optional flow chart of the device for predicting the interaction between an antibody and an antigen provided in an embodiment of the present application. The device for predicting the interaction between an antibody and an antigen comprises the following steps 901 to 905:

[0196] The acquisition module 901 is used to acquire the antibody sequence information of the target antibody and the antigen sequence information of the target antigen.

[0197] The feature extraction module 902 is used to extract antibody sequence features from the antibody sequence information and to extract antigen sequence features from the antigen sequence information.

[0198] The bidirectional attention feature calculation module 903 is used to determine the respective bidirectional attention weights between the antibody sequence features and the antigen sequence features, and extract the bidirectional attention features between the antibody sequence features and the antigen sequence features based on the respective bidirectional attention weights.

[0199] The structural feature mapping module 904 is used to map the antibody sequence features to the spatial structure of the target antibody to obtain the antibody structural features.

[0200] The target prediction module 905 is used to fuse the bidirectional attention feature and the antibody structure feature to obtain the target feature, and determine the target prediction result of the interaction between the target antibody and the target antigen based on the target feature.

[0201] The method for predicting the interaction between an antibody and an antigen proposed in the present application and the related device, which obtains the antibody sequence information of the target antibody and the antigen sequence information of the target antigen; extracts the antibody sequence features in the antibody sequence information, and extracts the antigen sequence features in the antigen sequence information; determines the respective two-way attention weights between the antibody sequence features and the antigen sequence features, and extracts the two-way attention features between the antibody sequence features and the antigen sequence features based on the respective two-way attention weights; helps to more comprehensively understand the interaction between the target antibody and the target antigen by simultaneously considering the two-way attention features of the attention from the target antibody to the target antigen, and the attention from the target antigen to the target antibody; then, the antibody The antibody structure characteristics are mapped onto the spatial structure of the target antibody to obtain the antibody structure characteristics; the antibody structure characteristics can better reflect the functional characteristics of the target antibody, so that the prediction device can better analyze the interaction between the target antibody and the target antigen; then, the two-way attention characteristics and the antibody structure characteristics are fused to obtain the target characteristics, and the target prediction result of the interaction between the target antibody and the target antigen is determined based on the target characteristics; without inputting the spatial structure information related to the target antibody into the prediction device, the prediction device proposed in the present application can process the input antibody sequence information and antigen sequence information based on limited computing resources, thereby improving the prediction accuracy of the interaction between the target antibody and the target antigen.

[0202] The specific implementation of the prediction device is basically the same as the specific implementation of the above prediction method, and will not be repeated here.

[0203] The embodiment of the present application also provides an electronic device, the electronic device includes a memory and a processor, the memory stores a computer program, and the processor implements the above prediction method when executing the computer program. The electronic device can be any intelligent terminal including a tablet computer, a car computer, etc.

[0204] like Fig.12 As shown, Fig.12 : is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present application, the electronic device includes:

[0205] The processor 1001 may be implemented by a general-purpose CPU (Central Processing Unit), a microprocessor, an application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present application.

[0206] The memory 1002 can be implemented in the form of a read-only memory (ROM), a static storage device, a dynamic storage device, or a random access memory (RAM). The memory 1002 can store an operating system and other application programs. When the technical solution provided in the embodiment of this specification is implemented by software or firmware, the relevant program code is stored in the memory 1002, and the processor 1001 calls and executes the prediction method of the embodiment of this application;

[0207] Input / output interface 1003, used to implement information input and output;

[0208] The communication interface 1004 is used to realize the communication interaction between the device and other devices. The communication can be realized through a wired manner (such as USB, network cable, etc.) or a wireless manner (such as mobile network, WI FI, Bluetooth, etc.);

[0209] A bus 1005 , which transmits information between various components of the device (e.g., the processor 1001 , the memory 1002 , the input / output interface 1003 , and the communication interface 1004 );

[0210] The processor 1001 , the memory 1002 , the input / output interface 1003 and the communication interface 1004 are connected to each other in communication within the device via the bus 1005 .

[0211] An embodiment of the present application further provides a computer-readable storage medium, which stores a computer program, and the computer program implements the above-mentioned prediction method when executed by a processor.

[0212] The memory, as a non-transient computer-readable storage medium, can be used to store non-transient software programs and non-transient computer executable programs. In addition, the memory 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 embodiments, the memory may optionally include a memory remotely disposed relative to the processor, and these remote memories may be connected to the processor 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.

[0213] The embodiments described in the embodiments of the present application are intended to more clearly illustrate the technical solutions of the embodiments of the present application and do not constitute a limitation on the technical solutions provided in the embodiments of the present application. Those skilled in the art will appreciate that with the evolution of technology and the emergence of new application scenarios, the technical solutions provided in the embodiments of the present application are also applicable to similar technical problems.

[0214] Those skilled in the art will appreciate that the technical solutions shown in the figures do not constitute a limitation on the embodiments of the present application, and may include more or fewer steps than shown in the figures, or a combination of certain steps, or different steps.

[0215] The device embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separated, that is, they may be located in one place or distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0216] Those skilled in the art will appreciate that all or some of the steps in the methods disclosed above, and the functional modules / units in the systems and devices may be implemented as software, firmware, hardware, or a suitable combination thereof.

[0217] The terms "first", "second", "third", "fourth", etc. (if any) in the specification of the present application and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present application described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any of their variations are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device comprising a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0218] It should be understood that in the present application, "at least one (item)" means one or more, and "plurality" means two or more. "And / or" is used to describe the association relationship of associated objects, indicating that three relationships may exist. For example, "A and / or B" can mean: only A exists, only B exists, and A and B exist at the same time, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single or plural items. For example, at least one of a, b or c can mean: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, c can be single or multiple.

[0219] In the several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the above units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0220] The units described above as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0221] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0222] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application is essentially or the part that contributes to the prior art or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including multiple instructions to enable a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of various embodiments of the present application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (Read-Only Memory, referred to as ROM), random access memory (Random Access Memory, referred to as RAM), disk or optical disk and other media that can store programs.

[0223] The preferred embodiments of the present invention are described above with reference to the accompanying drawings, but the scope of the rights of the present invention is not limited thereto. Any modification, equivalent substitution and improvement made by a person skilled in the art without departing from the scope and essence of the present invention should be within the scope of the rights of the present invention.

Claims

1. A method for predicting the interaction between an antibody and an antigen, characterized in that: The method comprises: Obtaining antibody sequence information of the target antibody and antigen sequence information of the target antigen; Extracting antibody sequence features from the antibody sequence information, and extracting antigen sequence features from the antigen sequence information; Determine respective bidirectional attention weights between the antibody sequence features and the antigen sequence features, and extract bidirectional attention features between the antibody sequence features and the antigen sequence features based on the respective bidirectional attention weights; Mapping the antibody sequence characteristics onto the spatial structure of the target antibody to obtain antibody structural characteristics; The bidirectional attention feature and the antibody structural feature are integrated to obtain a target feature, and a target prediction result of the interaction between the target antibody and the target antigen is determined based on the target feature.

2. The method according to claim 1, characterized in that The determining of respective bidirectional attention weights between the antibody sequence feature and the antigen sequence feature, and extracting bidirectional attention features between the antibody sequence feature and the antigen sequence feature based on the respective bidirectional attention weights, comprises: According to the preset first group of parameters, respectively determine the first query information, the first key information and the first value information corresponding to the antigen sequence information; According to the preset second group parameters, respectively determine the second query information and the second key information corresponding to the antibody sequence information; Determining a first bidirectional attention weight characterizing the degree of attention of the target antigen to the target antibody based on the first query information and the second key information; Determining, based on the second query information and the first key information, a second bidirectional attention weight characterizing the degree of attention of the target antibody to the target antigen; The first value information is normalized according to the superimposed first bidirectional attention weight and the second bidirectional attention weight to obtain a bidirectional attention feature.

3. The method according to claim 1, characterized in that The target prediction result of determining the interaction between the target antibody and the target antigen based on the target feature comprises: The target prediction result is used to characterize at least one of a half-inhibitory concentration, an interaction category attribute, and a binding affinity index.

4. The method according to claim 1, characterized in that: The antibody sequence information is obtained by encoding and processing by an antibody sequence encoder, and the antigen sequence information is obtained by encoding and processing by an antigen sequence encoder; Before obtaining the antibody sequence information of the target antibody and the antigen sequence information of the target antigen, the method further includes: Obtaining a first protein language model and a second protein language model trained by universal protein data, and obtaining sample antibody data and sample antigen data; Training the first protein language model based on the sample antibody data until a preset first training condition is reached, thereby obtaining the antibody sequence encoder; The second protein language model is trained based on the sample antigen data until a preset second training condition is reached to obtain the antigen sequence encoder.

5. The method according to claim 4, characterized in that The antibody structural features are obtained by mapping the antibody structure encoder; The antibody structure encoder is trained by the following steps, comprising: Obtaining sample sequence information and sample structure information corresponding to the sample antibody; Inputting the sample sequence information into the antibody structure encoder, and inputting the sample structure information into a preset teacher network model; Mapping the sample sequence information through the antibody structure encoder to obtain student structure features, and mapping the sample structure information through the teacher network model to obtain teacher structure features; A feature loss value is calculated based on the teacher structure feature and the student structure feature, and the antibody structure encoder is trained based on the feature loss value to obtain the trained antibody structure encoder.

6. The method according to claim 5, characterized in that The feature loss value is calculated based on the teacher structure feature and the student structure feature, including: Calculating the cosine similarity between the teacher structural features and the student structural features; An exponential value corresponding to the cosine similarity is calculated based on preset training parameters, and a logarithmic operation result of the exponential value is determined as the feature loss value.

7. The method according to claim 5, characterized in that The target prediction result is obtained by prediction by a prediction model, wherein the prediction model comprises a trained antibody sequence encoder, a trained antigen sequence encoder and a trained antibody structure encoder; The prediction model is trained by the following steps, which include: Obtaining first sequence information of sample antibodies and second sequence information of sample antigens, and verifying the prediction results; extracting a first sequence feature from the first sequence information, and extracting a second sequence feature from the second sequence information; Determine a sample bidirectional attention weight between the first sequence feature and the second sequence feature, respectively, and extract a sample bidirectional attention feature between the first sequence feature and the second sequence feature based on each of the sample bidirectional attention weights; Mapping the first sequence feature to the spatial structure of the sample antibody to obtain a first structural feature; fusing the sample bidirectional attention feature and the first structural feature to obtain a sample feature, and determining a sample prediction result of the interaction between the sample antibody and the sample antigen based on the sample feature; A model loss value is calculated based on the verification prediction result and the sample prediction result, the prediction model is trained according to the model loss value, and the trained prediction model is obtained.

8. The method according to claim 7, characterized in that The calculating the model loss value based on the verification prediction result and the sample prediction result includes: Calculating a first difference between a preset unit result and the verification prediction result, and calculating a second difference between a preset unit result and the sample prediction result; Calculating a first logarithmic operation value of the sample prediction result, and determining a product of the verification prediction result and the first logarithmic operation value as a first multiplication value; Calculate a second logarithmic operation value of the second difference, and determine a product of the first difference and the second logarithmic operation value as a second product value; The model loss value is obtained based on the first multiplication value and the second multiplication value.

9. A prediction device for the interaction between an antibody and an antigen, characterized in that: The device comprises: An acquisition module, used to acquire the antibody sequence information of the target antibody and the antigen sequence information of the target antigen; A feature extraction module, used to extract antibody sequence features from the antibody sequence information and to extract antigen sequence features from the antigen sequence information; a two-way attention feature calculation module, used to determine the respective two-way attention weights between the antibody sequence feature and the antigen sequence feature, and extract the two-way attention feature between the antibody sequence feature and the antigen sequence feature based on the respective two-way attention weights; A structural feature mapping module, used to map the antibody sequence features to the spatial structure of the target antibody to obtain antibody structural features; A target prediction module is used to fuse the bidirectional attention feature and the antibody structure feature to obtain a target feature, and determine a target prediction result of the interaction between the target antibody and the target antigen based on the target feature.

10. An electronic device, characterized in that: The electronic device comprises a memory and a processor, the memory stores a computer program, and the processor implements the method for predicting the interaction between an antibody and an antigen according to any one of claims 1 to 8 when executing the computer program.

11. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for predicting the interaction between the antibody and the antigen according to any one of claims 1 to 8 is implemented.

Images