Method, apparatus, device and storage medium for determining antigen specificity

Abstract

The embodiment of the application provides a kind of antigen specificity determination method, device, equipment and storage medium, at least be applied to artificial intelligence field, medical field and adaptive immune receptor, wherein, method includes: double-chain biological information of cell receptor is carried out word encoding processing, obtains amino acid word sequence;Wherein, at least one amino acid word in the amino acid word sequence is indicated;Using pre-trained amino acid sequence prediction model, based on the amino acid word sequence, the cell receptor is extracted, and the amino acid sequence representation of the cell receptor is obtained;Wherein, the amino acid sequence prediction model is obtained by training the data obtained by shielding processing part of sample amino acid word representation in sample data;Based on the amino acid sequence representation, the antigen specificity of the cell receptor is determined;Through the application, the cell receptor can be accurately extracted, so that the antigen specificity of the cell receptor is accurately determined.

Description

TECHNICAL FIELD

[0001] Embodiments of the present application relate to the field of artificial intelligence technology, and relate to, but are not limited to, an antigen specificity determination method, device, equipment and storage medium. BACKGROUND

[0002] The human immune system is composed of innate immunity and adaptive immunity. Adaptive immunity is a kind of immune response that can recognize and initiate against specific pathogens (antigens) after contacting with specific pathogens. T cells and B cells are important components of the adaptive immune system. Antigen recognition is one of the key factors of T cell and B cell mediated immunity. T cells and B cells mainly interact with antigens through T cell receptors (TCR) and B cell receptors (BCR), respectively. TCR recognizes and binds to antigens presented by the major histocompatibility complex (MHC) on the cell membrane, while BCR directly binds to specific antigens. The complementarity determining regions (CDRs) on TCR and BCR recognize and specifically bind to antigen peptide molecules. Studies have shown that sequencing BCR or TCR can be used to diagnose B or T lymphocyte malignancies and analyze the effect after treatment. Therefore, predicting the antigen specificity of TCR and BCR will have a revolutionary impact on the treatment of infectious diseases and autoimmune diseases, as well as cancer immunovaccine design and other frontier research fields.

[0003] In related technologies, predicting the antigen specificity of TCR and BCR mainly includes methods based on artificial defined features and methods based on artificial intelligence such as machine learning / deep learning. Among them, the method based on artificial defined features mainly classifies or clusters TCR and BCR sequences based on artificial defined features. The method based on artificial intelligence usually automatically extracts and learns features based on TCR and BCR information to predict the antigen specificity of TCR and BCR.

[0004] However, the method based on artificial defined features cannot fully represent the useful information of TCR and BCR sequences and accurately depict the distance difference between sequences. The method based on artificial intelligence needs a large amount of known antigen binding specificity data for training to get better prediction results. When the data is less, the prediction result is often poor, and at the same time, the method in related technologies cannot comprehensively and comprehensively represent the influence of the encoding of the double-stranded pair of TCR and BCR pair on each other during encoding processing. Therefore, the method in related technologies cannot accurately predict the antigen specificity of TCR and BCR. SUMMARY

[0005] The embodiment of the present application provides an antigen specificity method, device, equipment and storage medium, at least applied to the field of artificial intelligence and the medical field, can carry out accurate word coding processing to double-chain biological information of cell receptor, and carries out accurate feature extraction to cell receptor based on amino acid word sequence after word coding processing, thereby accurately determining the antigen specificity of cell receptor.

[0006] The technical scheme of the embodiment of the present application is as follows:

[0007] The embodiment of the present application provides an antigen specificity determination method, which comprises the following steps:

[0008] The double-chain biological information of the cell receptor is subjected to word coding processing to obtain an amino acid word sequence; wherein the amino acid word sequence comprises at least one amino acid word representation; a pre-trained amino acid sequence prediction model is used to extract features of the cell receptor based on the amino acid word sequence, to obtain an amino acid sequence representation of the cell receptor; wherein the amino acid sequence prediction model is obtained by training data obtained by occlusion processing on part of sample amino acid word representations in sample data; and the antigen specificity of the cell receptor is determined based on the amino acid sequence representation.

[0009] The embodiment of the present application provides an antigen specificity determination device, which comprises: a word coding module, configured to perform word coding processing on double-chain biological information of a cell receptor to obtain an amino acid word sequence; wherein the amino acid word sequence comprises at least one amino acid word representation; a feature extraction module, configured to use a pre-trained amino acid sequence prediction model to extract features of the cell receptor based on the amino acid word sequence; wherein the amino acid sequence prediction model is obtained by training data obtained by occlusion processing on part of sample amino acid word representations in sample data; and a determination module, configured to determine the antigen specificity of the cell receptor based on the amino acid sequence representation.

[0010] In some embodiments, the word coding module is further configured to: obtain two peptide chains of the cell receptor, and determine double-chain biological information of the cell receptor based on the two peptide chains; determine a monomer unit quantity corresponding to the word coding processing; encode consecutive amino acids in the double-chain biological information of the cell receptor as one amino acid word representation per monomer unit quantity to form the amino acid word sequence; wherein each two adjacent amino acid word representations have a preset number of overlapping amino acids, and the preset number is less than the monomer unit quantity by 1.

[0011] In some embodiments, the cell receptor comprises a T cell receptor or a B cell receptor; when the cell receptor is a T cell receptor, the two peptide chains comprise an alpha chain and a beta chain; when the cell receptor is a B cell receptor, the two peptide chains comprise a heavy chain and a light chain; and the number of monomer units is 3.

[0012] In some embodiments, the double-chain biological information comprises amino acid information corresponding to each of the two peptide chains; and the word encoding module is further configured to: encode each number of consecutive amino acids in each peptide chain into an amino acid word representation to form an amino acid subsequence corresponding to each peptide chain; and determine the amino acid word sequence according to the amino acid subsequences corresponding to the two peptide chains.

[0013] In some embodiments, the word encoding module is further configured to: splice the amino acid subsequences corresponding to the two peptide chains to obtain a spliced word sequence; and sequentially perform marking processing, segmentation processing, and position encoding processing on the spliced word sequence to obtain the amino acid word sequence.

[0014] In some embodiments, the feature extraction module is further configured to: perform feature extraction on the cell receptor based on the amino acid word sequence by using a pre-trained amino acid sequence prediction model to obtain an amino acid sequence representation of the cell receptor.

[0015] In some embodiments, the apparatus further comprises a model training module configured to train the amino acid sequence prediction model; wherein the model training module is further configured to: perform data preprocessing on obtained pre-training data to obtain sample data; wherein the sample data comprises sample double-chain biological information of a sample cell receptor; input the sample double-chain biological information into the amino acid sequence prediction model; perform word encoding processing on the sample double-chain biological information by using a word encoding processing layer of the amino acid sequence prediction model to obtain a sample amino acid word sequence; wherein the sample amino acid word sequence comprises at least one sample amino acid word representation; perform occlusion processing on the at least one sample amino acid word representation in the sample amino acid word sequence by using an occlusion processing layer of the amino acid sequence prediction model to obtain an occluded sample amino acid word sequence; perform amino acid sequence prediction on the sample cell receptor based on the occluded sample amino acid word sequence by using a prediction processing layer of the amino acid sequence prediction model to determine the sample amino acid word representation that is occluded during the occlusion processing, and obtain a sample amino acid sequence representation of the sample cell receptor; input the sample amino acid sequence representation into a preset loss model to obtain a loss result; and correct model parameters in the word encoding processing layer, the occlusion processing layer, and the prediction processing layer based on the loss result to obtain a trained amino acid sequence prediction model.

[0016] In some embodiments, the determining the antigen specificity of the cell receptor is achieved by a multilayer perceptron; the model training module is further configured to: obtain fine-tuning sample data, the fine-tuning sample data including unoccluded double-chain sample biological information and epitope information of a sample cell receptor; fine-tune the model parameters in the trained amino acid sequence prediction model and the multilayer perceptron by taking the epitope information as label data, to obtain a fine-tuned amino acid sequence prediction model and a fine-tuned multilayer perceptron; accordingly, the feature extraction is further configured to extract features of the cell receptor by using the fine-tuned amino acid sequence prediction model, and the determining module is further configured to determine the antigen specificity of the cell receptor by using the fine-tuned multilayer perceptron.

[0017] In some embodiments, the model training module is further configured to: determine, from the sample amino acid word sequence, a randomly selected at least one sample amino acid word representation as a target amino acid word representation; determine adjacent amino acid word representations adjacent to the target amino acid word representation; wherein the adjacent amino acid word representations include a first adjacent amino acid word representation adjacent to a first side of the target amino acid word representation, and a second adjacent amino acid word representation adjacent to a second side of the target amino acid word representation; when the target amino acid word representation is located at a sequence start position of the amino acid word sequence, the adjacent amino acid word representations include a second adjacent amino acid word representation adjacent to a second side of the target amino acid word representation; when the target amino acid word representation is located at a sequence end position of the amino acid word sequence, the adjacent amino acid word representations include a first adjacent amino acid word representation adjacent to a first side of the target amino acid word representation; and perform occlusion processing on the target amino acid word representation, overlapping amino acids in the first adjacent amino acid word representation, and overlapping amino acids in the second adjacent amino acid word representation, to obtain an occluded sample amino acid word sequence.

[0018] In some embodiments, the model training module is further configured to: input the sample amino acid sequence representation and the sample double-chain biological information into the preset loss model; determine, by a cross-entropy loss function in the preset loss model, a sequence distance between the sample amino acid sequence representation and the sample double-chain biological information; and determine the loss result according to the sequence distance.

[0019] In some embodiments, the amino acid sequence representation is a multi-modal feature; and the determining module is further configured to: input the amino acid sequence into the multilayer perceptron; and perform mapping processing on the multi-modal feature corresponding to the amino acid sequence by the multilayer perceptron, to obtain the antigen specificity of the cell receptor.

[0020] In some embodiments, the model training module is further configured to: filter data belonging to a specific object from the pre-training data; perform double-chain data pairing analysis on the data belonging to the specific object to determine double-chain pairing data, wherein the double-chain pairing data refers to data of two peptide chains having a pairing; perform data length analysis on the double-chain pairing data to obtain data length of each double-chain pairing data; and determine double-chain pairing data having a data length less than a length threshold as the sample data.

[0021] The embodiment of the present application provides an antigen specificity determination device, comprising:

[0022] The memory is configured to store executable instructions, and the processor is configured to execute the executable instructions stored in the memory to implement the antigen specificity determination method.

[0023] The embodiment of the present application provides a computer program product or a computer program, the computer program product or the computer program comprising executable instructions stored in a computer readable storage medium; wherein the processor of the antigen specificity determination device reads the executable instructions from the computer readable storage medium and executes the executable instructions to implement the antigen specificity determination method.

[0024] The embodiment of the present application provides a computer readable storage medium storing executable instructions, for causing the processor to execute the executable instructions to implement the antigen specificity determination method.

[0025] The embodiment of the present application has the following beneficial effects: the double-chain biological information of the cell receptor is subjected to word encoding processing to obtain an amino acid word sequence; then a pre-trained amino acid sequence prediction model is used to extract features of the cell receptor based on the amino acid word sequence to obtain an accurate amino acid sequence representation of the cell receptor; and thus the antigen specificity of the cell receptor can be determined based on the amino acid sequence representation. In the embodiment of the present application, since the double-chain biological information of the cell receptor is subjected to word encoding processing, the influence of the paired double-chain of the cell receptor on each other can be comprehensively and integrally represented, so that accurate word encoding processing can be performed on the gene information of the cell receptor, and since the amino acid sequence prediction model is trained by using data obtained by performing occlusion processing on part of the sample amino acid word representation in the sample data, the amino acid sequence prediction model can be prevented from converging too quickly, thereby further improving the accuracy of feature extraction of the cell receptor and ensuring the accuracy of the determined antigen specificity of the cell receptor. BRIEF DESCRIPTION OF DRAWINGS

[0026] Figure 1 is an optional architecture schematic diagram of the antigen specificity determination system provided by the embodiment of the present application;

[0027] Figure 2 is a structural schematic diagram of an antigen specificity determination device provided by an embodiment of the present application;

[0028] Figure 3 is an optional flowchart of an antigen specificity determination method provided by an embodiment of the present application;

[0029] Figure 4 is another optional flowchart of an antigen specificity determination method provided by an embodiment of the present application;

[0030] Figure 5 is an implementation flowchart of a training method of an amino acid sequence prediction model provided by an embodiment of the present application;

[0031] Figure 6 is an implementation flowchart of a data preprocessing method provided by an embodiment of the present application;

[0032] Figure 7 is a model framework schematic diagram of an algorithm corresponding to an antigen specificity determination method provided by an embodiment of the present application;

[0033] Figure 8 is a blocking effect schematic diagram before a blocking strategy is modified provided by an embodiment of the present application;

[0034] Figure 9 is a blocking effect schematic diagram after a blocking strategy is modified provided by an embodiment of the present application;

[0035] Figure 10 is a performance comparison schematic diagram of an embodiment of the present application and a method in the related art on TCR antigen specificity recognition on a public data set;

[0036] Figure 11 is a performance comparison schematic diagram of an embodiment of the present application and a single chain method on TCR antigen specificity recognition on a public data set;

[0037] Figure 12 is a performance comparison schematic diagram of an embodiment of the present application and other encoding methods on TCR antigen specificity recognition on a public data set. DETAILED DESCRIPTION

[0038] In order to make the purposes, technical solutions and advantages of the present application clearer, the present application will be further described in detail below with reference to the drawings, and the described embodiments should not be regarded as limiting the present application. All other embodiments obtained by a person of ordinary skill in the art without making any creative labor shall fall within the scope of protection of the present application.

[0039] In the following description, "some embodiments" are referred to, which describe a subset of all possible embodiments, but it can be understood that "some embodiments" can be the same subset or different subsets of all possible embodiments, and can be combined with each other without conflict. Unless otherwise defined, all technical and scientific terms used in the embodiments of the present application are the same as the meanings commonly understood by those skilled in the art to which the embodiments of the present application belong. The terms used in the embodiments of the present application are only for the purpose of describing the embodiments of the present application, and are not intended to limit the present application.

[0040] Before explaining the antigen specificity determination method of the embodiments of the present application, the methods in the related art are first described.

[0041] In the related art, the methods for predicting the antigen specificity of TCR and BCR mainly include methods based on artificially defined features and methods based on artificial intelligence such as machine learning / deep learning. Among them, the methods based on artificial definition include GLIPH (a software for functional clustering of TCR sequences), TCRdist (a tool box supporting python API, used for analyzing the instruction set of TCR), etc. Such methods mainly classify or cluster TCR and BCR sequences based on artificially defined features. For example, TCRdist uses weighted Hamming distance to measure the distance between different cell receptors, which is used for clustering TCR. The methods based on artificial intelligence include DeepTCR (a python package with an unsupervised and supervised collection of deep learning methods for analyzing T cell receptor sequencing data), TCRAI, TCR-BERT (a model based on Transformer), etc. Such methods automatically extract and learn features based on TCR and BCR sequence information to predict the antigen specificity of TCR and BCR.

[0042] However, for the methods based on artificial definition, such as GLIPH, TCRdist, etc., the artificially defined Hamming distance and other features have non-learnability and limitations, and may not be able to fully represent the useful information of TCR / BCR sequences and accurately depict the distance difference between sequences. For the methods based on artificial intelligence, such as DeepTCR, TCRAI, etc., a large amount of known antigen binding specificity data is needed for training to obtain better prediction results. When the data is less, the prediction result is often poor. And the TCR-BERT method uses TCR single-chain data for pre-training, without considering the information of double-chain in the pre-training stage, so the encoding may not be able to comprehensively and comprehensively represent the influence of TCR / BCR paired double-chain on each other. And it is only a method for TCR, without considering BCR.

[0043] Based on the problems existing in the related art, the antigen specificity determination method of the embodiments of the present application is based on the pre-trained language representation model (BERT, Bidirectional Encoder Representation from Transformers) of natural language processing to extract and encode the sequence information of TCR / BCR for TCR / BCR antigen specificity recognition. The embodiments of the present application are the first time to use double-chain TCR / BCR data for BERT model pre-training, and encode the paired CDR3 sequences.

[0044] In the antigen specificity determination method provided by the embodiments of the present application, first, the double-chain biological information of the cell receptor is word encoded to obtain an amino acid word sequence; wherein the amino acid word sequence includes at least one amino acid word representation; then, a pre-trained amino acid sequence prediction model is used to extract features of the cell receptor based on the amino acid word sequence to obtain an amino acid sequence representation of the cell receptor; wherein the amino acid sequence prediction model is obtained by training data obtained by occluding part of the sample amino acid word representation in the sample data; finally, the antigen specificity of the cell receptor is determined based on the amino acid sequence representation. In this way, since the double-chain biological information of the cell receptor is word encoded, the influence of the paired double-chain of the cell receptor on each other can be comprehensively and comprehensively represented, so that the gene information of the cell receptor can be accurately word encoded, and since the amino acid sequence prediction model is obtained by training the data obtained by occluding part of the sample amino acid word representation in the sample data, the amino acid sequence prediction model can be prevented from converging too quickly, thereby further improving the accuracy of feature extraction of the cell receptor and ensuring the accuracy of the determined antigen specificity of the cell receptor.

[0045] The antigen specificity determination device provided in the embodiments of the present application can be implemented as a terminal or a server. In one implementation, the antigen specificity determination device provided in the embodiments of the present application can be implemented as any terminal with a data processing function, such as a notebook computer, a tablet computer, a desktop computer, a mobile device (for example, a mobile phone, a portable music player, a personal digital assistant, a dedicated message device, a portable game device), a smart robot, a smart home appliance, and a smart vehicle device; in another implementation, the antigen specificity determination device provided in the embodiments of the present application can also be implemented as a server, where the server can be a stand-alone physical server, a server cluster or a distributed system composed of multiple physical servers, a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN, Content Delivery Network), and basic cloud computing services such as big data and artificial intelligence platforms. The terminal and the server can be directly or indirectly connected through wired or wireless communication, which is not limited in the embodiments of the present application. In the following, an exemplary application of the antigen specificity determination device implemented as a server will be described.

[0046] Referring to Figure 1 , Figure 1 is an optional architecture schematic diagram of the antigen specificity determination system provided in the embodiments of the present application. The embodiments of the present application take the antigen specificity prediction task of the antigen specificity determination method applied to any one of TCR and BCR as an example for description, that is, the embodiments of the present application can be applied to the research and development and experimental stage of the biomedical field, or the treatment field of infectious diseases and autoimmune diseases, and the cancer immunovaccine design field, to predict the antigen specificity of the cell receptor. In the embodiments of the present application, an antigen specificity prediction platform of a cell receptor can be provided, and the antigen specificity of the cell receptor can be predicted in the prediction platform. In the implementation process, the prediction platform can be implemented as an antigen specificity prediction application. To support the antigen specificity prediction application, the terminal of the embodiments of the present application is at least installed with the antigen specificity prediction application, and the antigen specificity of the cell receptor can be determined by using the antigen specificity determination method provided in the embodiments of the present application through the antigen specificity prediction application.

[0047] In the embodiment of the present application, the antigen specificity determination system 10 at least includes a terminal 100, a network 200 and a server 300, wherein the server 300 is a server of an antigen specificity prediction application. The server 300 can constitute an antigen specificity determination device of the embodiment of the present application. The terminal 100 connects the server 300 through the network 200, and the network 200 can be a wide area network or a local area network, or a combination of the two. When running the antigen specificity prediction application, the terminal 100 obtains the double-chain biological information of the cell receptor through the client of the antigen specificity prediction application, and encapsulates the double-chain biological information of the cell receptor into an antigen specificity determination request, and sends the antigen specificity determination request to the server 300. The server 300 parses the double-chain biological information from the antigen specificity determination request, and performs word encoding processing on the double-chain biological information of the cell receptor to obtain an amino acid word sequence; wherein the amino acid word sequence includes at least one amino acid word representation; then, a pre-trained amino acid sequence prediction model is used to extract features of the cell receptor based on the amino acid word sequence to obtain an amino acid sequence representation of the cell receptor; wherein the amino acid sequence prediction model is obtained by training data obtained by occlusion processing on part of the sample amino acid word representations in the sample data; finally, the antigen specificity of the cell receptor is determined based on the amino acid sequence representation. After obtaining the antigen specificity of the cell receptor, the server 300 sends the antigen specificity of the cell receptor to the terminal 100 for researchers to study and refer.

[0048] In some embodiments, the antigen specificity determination method can also be implemented by the terminal 100, that is, the terminal serves as the execution subject to obtain the double-chain biological information of the cell receptor; perform word encoding processing on the double-chain biological information of the cell receptor to obtain an amino acid word sequence; then, the terminal performs amino acid sequence prediction on the cell receptor based on the amino acid word sequence to obtain an amino acid sequence representation of the cell receptor; finally, the terminal determines the antigen specificity of the cell receptor based on the amino acid sequence representation.

[0049] The antigen specificity determination method provided in the embodiment of the present application can also be implemented based on a cloud platform and through cloud technology, for example, the server 300 described above can be a cloud server. The double-chain biological information of the cell receptor is processed by the cloud server, or the cloud server performs amino acid sequence prediction on the cell receptor based on the amino acid word sequence, or the cloud server determines the antigen specificity of the cell receptor based on the amino acid sequence representation, etc.

[0050] In some embodiments, a cloud storage can also be provided, and the double-stranded biological information of the cell receptor can be stored in the cloud storage, or the amino acid word sequence can also be stored in the cloud storage, or the predicted amino acid sequence representation can also be stored in the cloud storage. In this way, when determining the antigen specificity of the cell receptor, the amino acid sequence representation can be obtained from the cloud storage, and then the antigen specificity of the cell receptor can be directly determined.

[0051] It should be noted that the cloud technology refers to a hosting technology that unifies a series of resources such as hardware, software, network, etc. in a wide area network or a local area network to realize data calculation, storage, processing and sharing. The cloud technology is a general term of network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model application, which can form a resource pool, and can be used on demand, and is flexible and convenient. Cloud computing technology will become an important support. The background service of the technical network system needs a large amount of calculation and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, every item may have its own identification mark in the future, and needs to be transmitted to the background system for logical processing. Different levels of data will be processed separately, and various industry data need strong system support, which can only be realized through cloud computing.

[0052] Figure 2 is a structural schematic diagram of an antigen specificity determination device provided by the embodiment of the present application, Figure 2 The antigen specificity determination device shown includes at least one processor 310, a memory 350, at least one network interface 320 and a user interface 330. The various components in the antigen specificity determination device are coupled together through a bus system 340. It can be understood that the bus system 340 is used to realize the connection communication between the components. In addition to including a data bus, the bus system 340 also includes a power bus, a control bus and a status signal bus. However, for the purpose of clear illustration, all kinds of buses are marked as the bus system 340 in Figure 2 .

[0053] The processor 310 can be an integrated circuit chip with signal processing capability, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor.

[0054] The user interface 330 includes one or more output devices 331 that enable the presentation of media content, and one or more input devices 332.

[0055] The memory 350 can be removable, non-removable, or a combination thereof. Exemplary hardware devices include solid-state memory, hard drives, optical drives, etc. The memory 350 optionally includes one or more storage devices remotely located from the processor(s) 310. The memory 350 comprises a volatile memory or non-volatile memory, and can also include both volatile and non-volatile memory. Non-volatile memory can be read only memory (ROM), volatile memory can be random access memory (RAM). The memory 350 described in embodiments of the present application is intended to include any suitable type of memory. In some embodiments, the memory 350 is capable of storing data to support various operations, examples of which include programs, modules, and data structures or a subset or superset thereof, which are described below.

[0056] The operating system 351 includes a system program for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks;

[0057] The network communication module 352 is used to reach other computing devices via one or more (wired or wireless) network interfaces 320, exemplary network interfaces 320 include Bluetooth, wireless compatibility certification (WiFi), and universal serial bus (USB), etc.

[0058] The input processing module 353 is used to detect and translate one or more user inputs or interactions from one or more input devices 332.

[0059] In some embodiments, the apparatus provided by embodiments of the present application can be implemented in software, Figure 2 An antigen specificity determination apparatus 354 stored in the memory 350 is shown, which can be an antigen specificity determination apparatus in an antigen specificity determination device, which can be software in the form of programs and plug-ins, etc., including the following software modules: word encoding module 3541, feature extraction module 3542, and determination module 3543, which are logical, so any combination or further splitting can be made according to the implemented functions. The functions of each module will be described below.

[0060] In some embodiments, the apparatus provided by the embodiments of the present application can be implemented in a hardware manner. For example, the apparatus provided by the embodiments of the present application can be a processor in the form of a hardware decoding processor, which is programmed to perform the antigen specificity determination method provided by the embodiments of the present application. For example, the processor in the form of a hardware decoding processor can be implemented by using one or more application specific integrated circuits (ASICs), DSPs, programmable logic devices (PLDs), complex programmable logic devices (CPLDs), field programmable gate arrays (FPGAs), or other electronic elements.

[0061] The antigen specificity determination method provided by the embodiments of the present application can be executed by an antigen specificity determination device. The antigen specificity determination device can be any terminal with a data processing function, or can be a server. That is, the antigen specificity determination method provided by the embodiments of the present application can be executed by a terminal, can be executed by a server, or can be executed by interaction between a terminal and a server.

[0062] Referring to Figure 3 , Figure 3 is an optional flowchart of the antigen specificity determination method provided by the embodiments of the present application. The following will be described with reference to the steps shown in Figure 3 It should be noted that the antigen specificity determination method in Figure 3 is described by taking a server as an execution subject.

[0063] In step S301, the double-stranded biological information of the cell receptor is subjected to word encoding processing to obtain an amino acid word sequence. The amino acid word sequence includes at least one amino acid word.

[0064] Here, the cell receptor can be an adaptive immune receptor, and the cell receptor includes a T cell receptor (TCR) or a B cell receptor (BCR).

[0065] TCR is a characteristic marker on the surface of all T cells, combined with CD3 by non-covalent bond to form TCR-CD3 complex; the role of TCR is to recognize antigen. TCR is a heterodimer composed of two different peptide chains, composed of α chain and β chain two peptide chains, each peptide chain can be divided into variable region (V region), constant region (C region), transmembrane region and cytoplasmic region and so on. TCR molecule belongs to immunoglobulin superfamily, and the antigen specificity of TCR exists in V region; V region has three hypervariable regions CDR1, CDR2 and CDR3, among which CDR3 has the largest variation, which directly determines the antigen binding specificity of TCR. When TCR recognizes MHC-antigen peptide complex, CDR1 and CDR2 recognize and combine with the side wall of MHC molecule antigen binding groove, and CDR3 directly combines with antigen peptide.

[0066] TCR is divided into two types: TCR1 and TCR2; TCR1 is composed of γ chain and δ chain two chains, and TCR2 is composed of α chain and β chain two chains. In the embodiments of the application, when the cell receptor is TCR1, the double-stranded biological information is the amino acid information carried on the γ chain and the δ chain; when the cell receptor is TCR2, the double-stranded biological information is the amino acid information carried on the α chain and the β chain. In one implementation mode, the double-stranded biological information can be the amino acid sequence information of the CDR3 region of TCR.

[0067] BCR is a molecule responsible for specific recognition and binding of antigen on the surface of B cells, and BCR is essentially a membrane surface immunoglobulin (mIg, membrane immunoglobulin). BCR has antigen binding specificity (i.e. antigen specificity), and the diversity of BCR of each individual is as high as 5x10 13 , which constitutes a huge BCR library, and gives the individual a huge potential to recognize various antigens and produce specific antibodies. Two types of membrane surface immunoglobulins, mIg M and mIg D, are expressed on the surface of B cells for specific recognition and binding of antigen. The mIg is connected by two heavy chains and two light chains. Among them, the heavy chain is divided into variable region (V region, about 110 amino acid residues), constant region (C region, about 330 amino acid residues), transmembrane region (26 amino acid residues) and cytoplasmic region (3 amino acid residues); and the light chain only has V region and C region. The V region of the heavy chain and the light chain has three regions with highly variable amino acid composition and arrangement order, which can form a complementary spatial conformation with the epitope of the antigen, called the complementarity determining region (CDR). The three CDRs on the mIg are involved in the recognition of the antigen, which determines the antigen specificity of the BCR together. In the embodiments of the application, when the cell receptor is BCR, the double-stranded biological information is the amino acid information carried on the heavy chain and the light chain. In one implementation mode, the double-stranded biological information can be the amino acid sequence information of the CDR3 region of BCR.

[0068] The word encoding processing refers to a processing process of encoding double-stranded biological information into a sentence. Since there is no clear concept of a word in a biological sequence, a fixed number of amino acids in the double-stranded biological information can be selected as a word to form a sentence.

[0069] In the embodiments of the present application, when a fixed number of amino acids is selected to encode a word, the fixed number can be based on a preset number of monomer units, for example, k monomer units (kmer) can be selected as the number of monomer units to sequentially encode the amino acid sequence information in the double-stranded biological information.

[0070] In some embodiments, when the word encoding processing is performed, there is a certain number of amino acids overlapping between two adjacent amino acid word representations. Here, the number of overlapping amino acids can be the number of monomer units used in encoding minus 1. When the amino acid sequence information includes N amino acids, then through the word encoding processing, N-2 amino acid word representations can be obtained, which constitute an amino acid word sequence, i.e., the above-mentioned sentence. This sentence can be input as input text to the amino acid sequence prediction model in the subsequent antigen specificity determination process.

[0071] In step S302, a pre-trained amino acid sequence prediction model is used to extract features of the cell receptor based on the amino acid word sequence to obtain an amino acid sequence representation of the cell receptor.

[0072] In the embodiments of the present application, the feature extraction of the cell receptor can be the prediction of the amino acid sequence to determine the amino acid sequence representation corresponding to the double-stranded biological information. The amino acid sequence prediction model can be used to predict the amino acid sequence of the cell receptor. In the implementation process, the amino acid word sequence can be input as input data to the amino acid sequence prediction model, and the amino acid sequence representation of the cell receptor can be output by the amino acid sequence prediction model. The amino acid sequence prediction model is obtained by training the data obtained by occluding part of the sample amino acid word representation in the sample data.

[0073] Here, since the amino acid sequence prediction model is a pre-trained prediction model, the amino acid sequence representation of the cell receptor can be accurately predicted when the amino acid sequence representation is predicted. Moreover, when training the amino acid sequence prediction model, the sample data input will be occluded for part of the sample amino acid word representation, so that the amino acid sequence prediction model can predict the occluded part, realize the self-learning process of the occluded part based on the input sample data, so as to not only ensure that the model will not converge too quickly, but also accurately learn the parameters in the model.

[0074] In step S303, the antigen specificity of the cellular receptor is determined based on the amino acid sequence representation.

[0075] In the embodiments of the present application, after obtaining the amino acid sequence representation of the cellular receptor, the antigen specificity of the cellular receptor is determined based on the amino acid sequence representation of the cellular receptor.

[0076] Here, the antigen specificity is the most prominent feature of the immune response, and is the theoretical basis for immunological diagnosis and prevention. The antigen specificity refers to the property that an antigen can only bind to corresponding antibodies and sensitized lymphocytes. That is, the antigen specificity is reflected in immunogenicity and immunoreactivity. The antigen specificity reflected in immunogenicity refers to that a specific antigen only causes a specific immune response. The antigen specificity reflected in immunoreactivity refers to that a specific antigen only specifically binds to corresponding antibodies and / or sensitized lymphocytes to appear a reaction. The antigen specificity is first determined by the chemical composition, but even if the chemical composition is the same, the specificity is not the same due to the difference in the stereo configuration.

[0077] In the embodiments of the present application, after obtaining the amino acid sequence of the cellular receptor, the chemical composition and the stereo configuration of the cellular receptor are obtained. Therefore, the classification processing can be performed based on the chemical composition and the stereo configuration of the cellular receptor, so that the antigen specificity of the cellular receptor is mapped.

[0078] In the implementation process, the features corresponding to the amino acid sequence can be mapped (i.e., classified) by a classification head containing a multilayer perceptron (MLP, Multilayer Perceptron) to obtain the antigen specificity of the cellular receptor. In the embodiments of the present application, the multilayer perceptron can be a fully connected network.

[0079] The antigen specificity determination method provided in the embodiments of the present application performs word encoding processing on the double-chain biological information of the cell receptor to obtain an amino acid word sequence; then a pre-trained amino acid sequence prediction model is used to perform feature extraction on the cell receptor based on the amino acid word sequence to obtain an accurate amino acid sequence representation of the cell receptor; wherein the amino acid sequence prediction model is obtained by training data obtained by performing occlusion processing on part of the sample amino acid word representations in the sample data; thus, based on the amino acid sequence representation, the antigen specificity of the cell receptor can be determined. In the embodiments of the present application, since the double-chain biological information of the cell receptor is subjected to word encoding processing, the influence of the encoding of each other on the paired double-chain of the cell receptor can be comprehensively and comprehensively characterized, so that accurate word encoding processing can be performed on the gene information of the cell receptor, and since the amino acid sequence prediction model is obtained by training data obtained by performing occlusion processing on part of the sample amino acid word representations in the sample data, the prediction model can be prevented from converging too quickly during amino acid sequence prediction, thereby further improving the accuracy of amino acid sequence prediction of the cell receptor and ensuring the accuracy of the determined antigen specificity of the cell receptor.

[0080] In some embodiments, the antigen specificity determination method is applied to the antigen specificity prediction task of any one of TCR and BCR, that is, the antigen specificity determination method provided in the embodiments of the present application can be applied to the research and development and experimental stage of the biomedical field, or the treatment field of infectious diseases and autoimmune diseases, and the cancer immune vaccine design field to predict the antigen specificity of the cell receptor. In the embodiments of the present application, an antigen specificity prediction platform of a cell receptor can be provided, and the antigen specificity of the cell receptor can be predicted in the prediction platform. In the implementation process, the prediction platform can be implemented as an antigen specificity prediction application. The antigen specificity determination system at least includes a terminal and a server, and the terminal at least installs the antigen specificity prediction application, and the antigen specificity of the cell receptor can be determined by using the antigen specificity determination method provided in the embodiments of the present application through the antigen specificity prediction application. The server can be a server of the antigen specificity prediction application.

[0081] Figure 4 is another optional flowchart of the antigen specificity determination method provided in the embodiments of the present application, as shown in Figure 4 the method includes the following steps:

[0082] Step S401, the terminal acquires two peptide chains of the cell receptor through the client of the antigen specificity prediction application.

[0083] In the embodiments of the present application, the two peptide chains of the cell receptor can be obtained through experiments, and the peptide chain information corresponding to the two peptide chains is correspondingly obtained.

[0084] Step S402, the terminal encapsulates the two peptide chains of the cell receptor into the antigen specificity determination request.

[0085] Here, the peptide chain information corresponding to the two peptide chains can be encapsulated into the antigen specificity determination request

[0086] Step S403, the terminal sends the antigen specificity determination request to the server.

[0087] Step S404, the server parses the antigen specificity determination request to obtain the two peptide chains of the cell receptor.

[0088] In some embodiments, the cell receptor includes a T cell receptor or a B cell receptor; when the cell receptor is a T cell receptor, the two peptide chains include an alpha chain and a beta chain; when the cell receptor is a B cell receptor, the two peptide chains include a heavy chain and a light chain.

[0089] Step S405, the server determines the double-chain biological information of the cell receptor based on the two peptide chains.

[0090] In the embodiments of the present application, any sequencing method can be used to sequence the two peptide chains to obtain the double-chain biological information of the cell receptor.

[0091] Step S406, the server determines the number of monomer units corresponding to the word encoding processing.

[0092] Here, the number of monomer units refers to the number of amino acids included in each amino acid word representation during word encoding processing.

[0093] Step S407, the server encodes the continuous amino acids in the double-chain biological information of the cell receptor into one amino acid word representation for each number of monomer units to form an amino acid word sequence.

[0094] Each two adjacent amino acid word representations have a preset number of overlapping amino acids, and the preset number is one less than the number of monomer units. The amino acid word sequence includes at least one amino acid word representation.

[0095] In some embodiments, the number of monomer units can be 3, so each two adjacent amino acid word representations have two overlapping amino acids.

[0096] In some embodiments, the double-chain biological information includes gene information corresponding to each of the two peptide chains. In step S407, encoding the continuous amino acids in the double-chain biological information of the cell receptor into one amino acid word representation for each number of monomer units can be achieved by the following steps S4071 and S4072 (not shown in the figure):

[0097] In step S4071, the consecutive amino acids in each monomer unit in each peptide chain are encoded into an amino acid word representation, to form an amino acid subsequence corresponding to each peptide chain.

[0098] In step S4072, the amino acid word sequence is determined according to the amino acid subsequences corresponding to the two peptide chains.

[0099] Here, the amino acid subsequences corresponding to the two peptide chains can be spliced to obtain a spliced word sequence, and then the spliced word sequence can be sequentially subjected to marking processing, segmentation processing, and position encoding processing to obtain the amino acid word sequence.

[0100] In the embodiments of the present application, the splicing processing refers to connecting the two amino acid subsequences corresponding to the two peptide chains into an amino acid word sequence, where the sequence length of the amino acid word sequence is equal to the sum of the sequence lengths of the two amino acid subsequences. In the implementation process, if the cellular receptor is a T cell receptor, the two peptide chains include an a chain and a β chain, and thus the amino acid subsequences corresponding to the two peptide chains are an amino acid subsequence corresponding to the a chain and an amino acid subsequence corresponding to the β chain. In the splicing processing, the amino acid subsequence corresponding to the β chain can be spliced after the amino acid subsequence corresponding to the a chain to form an amino acid word sequence. If the cellular receptor is a B cell receptor, the two peptide chains include a light chain and a heavy chain, and thus the amino acid subsequences corresponding to the two peptide chains are an amino acid subsequence corresponding to the light chain and an amino acid subsequence corresponding to the heavy chain. In the splicing processing, the amino acid subsequence corresponding to the heavy chain can be spliced after the amino acid subsequence corresponding to the light chain to form an amino acid word sequence.

[0101] The marking processing refers to marking at least one amino acid word representation with a specific function and a specific meaning in the amino acid word sequence to obtain a marking identifier, so that the amino acid word representation can be determined based on the marking identifier in the subsequent prediction and processing. The segmentation processing refers to dividing the amino acid word sequence into multiple amino acid word sequence segments as a sentence, and the prediction can be performed based on the multiple amino acid word sequence segments in the subsequent amino acid sequence prediction process. The position encoding processing refers to encoding the position of the amino acid word representation in the amino acid word sequence to obtain position encoding information of each amino acid word representation in the amino acid word sequence. For example, when the position encoding processing is performed on the third amino acid word representation in the amino acid word sequence, the position “3” of the amino acid word representation can be encoded into the position encoding information of the amino acid word representation, so that the position information of the corresponding amino acid word representation in the amino acid word sequence can be obtained by decoding the position encoding information.

[0102] In step S408, the server performs feature extraction on the cellular receptor based on the amino acid word sequence to obtain an amino acid sequence representation of the cellular receptor.

[0103] In step S409, the server determines the antigen specificity of the cell receptor based on the amino acid sequence representation.

[0104] In step S410, the server sends the antigen specificity of the cell receptor to the terminal.

[0105] The antigen specificity determination method provided by the embodiments of the present application encodes the double-chain biological information of the cell receptor by word based on the number of monomer units, wherein each continuous amino acid of the number of monomer units is encoded into an amino acid word representation to form an amino acid word sequence, and each two adjacent amino acid word representations have a preset number of overlapping amino acids, and the preset number is less than the number of monomer units by 1. In this way, by continuous and overlapping encoding, the accuracy of the word encoding processing can be ensured by accurately encoding each amino acid in the double-chain biological information, and the information of any amino acid in the double-chain biological information is avoided to be missed, so as to ensure the accuracy of the word encoding processing.

[0106] In some embodiments, the amino acid sequence representation of the cell receptor can also be obtained by a pre-trained amino acid sequence prediction model based on the amino acid word sequence. The amino acid sequence prediction model includes a word encoding processing layer, a masking processing layer, and a prediction processing layer.

[0107] In the use process of the amino acid sequence prediction model, the word encoding processing layer is used to implement the word encoding processing of the double-chain biological information of the cell receptor in step S301, or to implement the word encoding processing in steps S405 to S407, so as to implement the word encoding processing of the double-chain biological information of the cell receptor to obtain the amino acid word sequence. The masking processing layer is used to implement the step of masking at least one sample amino acid word representation in the sample amino acid word sequence to obtain the masked sample amino acid word sequence. The prediction processing layer is used to implement the step of predicting the amino acid sequence of the cell receptor based on the amino acid word sequence in step S303 to obtain the amino acid sequence representation of the cell receptor.

[0108] In the training process of the amino acid sequence prediction model, the word encoding processing layer is used to implement the step of word encoding processing of the sample double-chain biological information of the sample cell receptor to obtain the sample amino acid word sequence. The masking processing layer is used to implement the step of masking at least one sample amino acid word representation in the sample amino acid word sequence to obtain the masked sample amino acid word sequence. The prediction processing layer is used to implement the step of predicting the amino acid sequence of the sample cell receptor based on the masked sample amino acid word sequence to obtain the sample amino acid sequence representation of the sample cell receptor.

[0109] The training process of the amino acid sequence prediction model is described below. Figure 5is a schematic diagram of an implementation process of a training method of an amino acid sequence prediction model provided by an embodiment of the present application, as shown in Figure 5 The method comprises the following steps:

[0110] In step S501, the pre-training data obtained is subjected to data preprocessing to obtain sample data.

[0111] Here, the sample data includes sample double-chain biological information of a sample cell receptor.

[0112] In step S502, the sample double-chain biological information is input into the amino acid sequence prediction model.

[0113] In step S503, the sample double-chain biological information is subjected to word encoding processing by a word encoding processing layer of the amino acid sequence prediction model to obtain a sample amino acid word sequence; wherein the sample amino acid word sequence includes at least one sample amino acid word representation.

[0114] In step S504, at least one sample amino acid word representation in the sample amino acid word sequence is subjected to mask processing by a mask processing layer of the amino acid sequence prediction model to obtain a sample amino acid word sequence after mask processing.

[0115] Here, the mask processing refers to masking part of the sample amino acid word representations in the sample amino acid word sequence, that is, before predicting the sample amino acid sequence representation, masking the information of the sample amino acid word representations in the positions before and after the current sample amino acid word representation in the sample amino acid word sequence.

[0116] In the embodiment of the present application, the amino acid sequence prediction model can be used to predict the amino acid sequence of the sample cell receptor based on the amino acid word sequence after mask processing to obtain the sample amino acid sequence representation of the sample cell receptor. The amino acid sequence prediction model can be an amino acid sequence prediction model based on BERT technology.

[0117] The reasons why part of the sample amino acid word representations need to be subjected to mask processing in the embodiment of the present application will be described below.

[0118] The reason why BERT needs mask processing is because of the use of the Transformer module, so to know why BERT needs mask is to ask why the Transformer module needs mask. But there is a difference between the two, because BERT only uses the encoding part of the Transformer module, without the decoding part, so compared with the two mask processing (key padding mask and attention mask) in the Transformer module, there is only key padding mask in BERT, that is, the information of the padding part is ignored, and the information after the current position also needs to be ignored in the decoding stage of the Transformer module, so attention mask is also needed. In short, the amino acid sequence prediction model performs part of the word mask processing when encoding the sentence during pre-training, which mainly functions to guess what the masked word is by using the words before and after the masked word, because the computer knows the correct value of the masked word, so it can also judge whether the word guessed by the amino acid sequence prediction model is accurate.

[0119] In the embodiments of the present application, during the mask processing, 15% of the sample amino acid word representations can be masked, and then the amino acid sequence prediction model can be used to predict the masked sample amino acid word representations. The amino acid sequence prediction model processes each sample amino acid word representation separately, that is, the word group information is not considered during the mask processing. For example, the sentence "The author of the poem is Li Bai" can be masked by the BERT model to obtain "The author of the poem is [Mask] Bai", and for the amino acid sequence prediction model of the embodiments of the present application, assuming that the sample amino acid word sequence is "CAVPGNNDMRF", the amino acid sequence prediction model can be masked to obtain "[Mask] [Mask] [Mask] PGN GNN NND NDM DMR M RF".

[0120] In the embodiments of the present application, by masking some word groups (i.e. sample amino acid word representations) in the sentence composed of sample amino acid word sequences and predicting the entire word group, the amino acid sequence prediction model can better capture the relationship between word groups and entities.

[0121] In some embodiments, the following steps S5041 to S5043 (not shown in the figure) can be used to mask part of the sample amino acid word representations in the sample amino acid word sequence:

[0122] Step S5041, from the sample amino acid word sequence, at least one randomly selected sample amino acid word representation is determined as a target amino acid word representation.

[0123] In the embodiments of the present application, the occlusion processing can be implemented in a random occlusion manner, that is, a certain number of sample amino acid word representations are randomly selected for occlusion processing.

[0124] In some embodiments, the occlusion ratio can be determined in advance or set in advance, and the occlusion ratio refers to the ratio between the number of determined target amino acid word representations and the total number of all sample amino acid word representations in the sample amino acid word sequence. Then, based on the occlusion ratio and the total number, the target number of target amino acid word representations is determined, and based on the determined data of the target amino acid word representations, the target number of target amino acid word representations is randomly determined from all sample amino acid word representations.

[0125] For example, the occlusion ratio can be set to 15%, so that 15% of the target amino acid word representations can be occluded during the occlusion processing.

[0126] In step S5042, adjacent amino acid word representations adjacent to the target amino acid word representation are determined.

[0127] Here, the adjacent amino acid word representations include a first adjacent amino acid word representation adjacent to the first side of the target amino acid word representation and a second adjacent amino acid word representation adjacent to the second side of the target amino acid word representation. The first side and the second side refer to the two opposite sides of the target amino acid word representation, for example, the first side can be the side of the amino acid word representation located before the target amino acid word representation, and the second side can be the side of the amino acid word representation located after the target amino acid word representation.

[0128] In some embodiments, when the target amino acid word representation is located at the sequence start position of the sample amino acid word sequence, the adjacent amino acid word representation includes the second adjacent amino acid word representation adjacent to the second side of the target amino acid word representation, and at this time, there is no first adjacent amino acid word representation adjacent to the first side of the target amino acid word representation; when the target amino acid word representation is located at the sequence end position of the sample amino acid word sequence, the adjacent amino acid word representation includes the first adjacent amino acid word representation adjacent to the first side of the target amino acid word representation, and at this time, there is no second adjacent amino acid word representation adjacent to the second side of the target amino acid word representation.

[0129] In step S5043, the overlapping amino acids in the target amino acid word representation, the first adjacent amino acid word representation, and the second adjacent amino acid word representation are occluded to obtain the occluded amino acid word sequence.

[0130] Here, the overlapping amino acids refer to the amino acid representation names of the overlapping amino acids in the two adjacent sample amino acid word representations. Since each sample amino acid word representation is composed of consecutive amino acid codes of the number of monomer units, there are a preset number of overlapping amino acids between each two adjacent sample amino acid word representations, and the preset number is one less than the number of monomer units.

[0131] For the target amino acid word representation and the first adjacent amino acid word representation, the overlapping amino acids refer to the preset number of amino acids in the first adjacent amino acid word representation close to the target amino acid word representation, or the preset number of amino acids in the target amino acid word representation close to the first adjacent amino acid word representation; that is, the preset number of amino acids at the rear of the first adjacent amino acid word representation, or the preset number of amino acids at the front of the target amino acid word representation.

[0132] For the target amino acid word representation and the second adjacent amino acid word representation, the overlapping amino acids refer to the preset number of amino acids in the second adjacent amino acid word representation close to the target amino acid word representation, or the preset number of amino acids in the target amino acid word representation close to the second adjacent amino acid word representation; that is, the preset number of amino acids at the front of the second adjacent amino acid word representation, or the preset number of amino acids at the rear of the target amino acid word representation.

[0133] For example, when the number of monomer units is 3 and the preset number is 2, for the target amino acid word representation and the first adjacent amino acid word representation, the overlapping amino acids refer to the last two amino acids in the first adjacent amino acid word representation, or the first two amino acids in the target amino acid word representation; for the target amino acid word representation and the second adjacent amino acid word representation, the overlapping amino acids refer to the first two amino acids in the second adjacent amino acid word representation, or the last two amino acids in the target amino acid word representation.

[0134] In the embodiments of the present application, since the occlusion processing is performed on part of the sample amino acid word representations after the word encoding processing, the amino acid sequence prediction model can be prevented from converging too quickly during the amino acid sequence prediction, thereby further improving the accuracy of the amino acid sequence prediction of the cell receptor.

[0135] In step S505, the amino acid sequence of the sample cell receptor is predicted based on the occluded sample amino acid word sequence by the prediction processing layer of the amino acid sequence prediction model, to determine the occluded sample amino acid word representation during the occlusion processing, and obtain the sample amino acid sequence representation of the sample cell receptor.

[0136] In some embodiments, after obtaining the sample amino acid sequence representation, the sample antigen specificity of the sample cell receptor can also be determined based on the sample amino acid sequence representation. Since the sample amino acid sequence representation is a multi-modal feature, the determination of the sample antigen specificity of the sample cell receptor can be implemented by a multi-layer perceptron.

[0137] Here, the sample amino acid sequence can be input into the multi-layer perceptron; then, the multi-modal feature corresponding to the sample amino acid sequence is mapped by the multi-layer perceptron to obtain the sample antigen specificity of the sample cell receptor.

[0138] In an implementation manner, the multi-layer perceptron can be a fully connected network, which can be connected after the amino acid sequence prediction model. After the amino acid sequence prediction model predicts the sample amino acid sequence representation of the sample cell receptor, the sample amino acid sequence representation is a sequence representation including a multi-modal feature. At this time, the sample amino acid sequence representation can be input into the multi-layer perceptron as an input feature of the multi-layer perceptron. The multi-modal feature corresponding to the sample amino acid sequence is mapped by the multi-layer perceptron to obtain the sample antigen specificity of the sample cell receptor.

[0139] Here, the mapping processing can be any type of classification processing, which can classify the antigen affinity of the sample cell receptor based on the multi-modal feature corresponding to the sample amino acid sequence to obtain the antigen affinity result corresponding to the sample cell receptor, so as to determine the sample antigen specificity of the sample cell receptor based on the antigen affinity result.

[0140] Step S506: inputting the sample amino acid sequence representation into a preset loss model to obtain a loss result.

[0141] In some embodiments, step S506 can be implemented by the following steps S5061 to S5064 (not shown in the figure):

[0142] Step S5061: inputting the sample amino acid sequence representation and the sample double-stranded biological information into the preset loss model.

[0143] Step S5062: determining the sequence distance between the sample amino acid sequence representation and the sample double-stranded biological information by a cross-entropy loss function in the preset loss model.

[0144] In the embodiments of the present application, the cross-entropy loss function is included in the preset loss model, and the end-to-end training of the amino acid sequence prediction model is implemented by the cross-entropy loss function. The sequence distance between the predicted sample amino acid sequence representation and the sample double-stranded biological information (i.e. the real amino acid sequence representation) can be calculated by the cross-entropy loss function.

[0145] Step S5063, determining the loss result according to the sequence distance.

[0146] Here, the greater the sequence distance, the smaller the similarity between the sample amino acid sequence representation and the sample double-stranded biological information; the smaller the sequence distance, the greater the similarity between the sample amino acid sequence representation and the sample double-stranded biological information.

[0147] In the embodiments of the present application, the sequence distance between the sample amino acid sequence representation and the sample double-stranded biological information is calculated by the loss function in the preset loss model, so as to obtain the loss result, which can accurately determine the difference between the predicted result and the true result of the amino acid sequence prediction model under the current model parameter.

[0148] Step S507, based on the loss result, correcting the model parameters in the word encoding processing layer, the occlusion processing layer and the prediction processing layer to obtain the trained amino acid sequence prediction model.

[0149] In the embodiments of the present application, the amino acid sequence prediction model is corrected based on the loss result, so as to realize the training of the amino acid sequence prediction model, obtain the amino acid sequence prediction model capable of accurately predicting the sample amino acid sequence representation, and determine the antigen specificity of the cell receptor based on the accurately predicted sample amino acid sequence representation.

[0150] In some embodiments, after the training of the amino acid sequence prediction model to obtain the trained amino acid sequence prediction model, the amino acid sequence prediction model can also be fine-tuned.

[0151] In the embodiments of the present application, a multi-layer perceptron can be connected after the amino acid sequence prediction model, which is used to determine the antigen specificity of the cell receptor. The trained amino acid sequence prediction model and the multi-layer perceptron can be fine-tuned based on the epitope information of the sample cell receptor, and only the model parameters in the trained amino acid sequence prediction model and the multi-layer perceptron are fine-tuned.

[0152] In the implementation process, fine-tuning sample data can be acquired, wherein the fine-tuning sample data includes unoccluded double-chain sample biological information and epitope information of a sample cell receptor; here, the unoccluded double-chain sample biological information is double-chain sample biological information of the sample cell receptor. Then, the epitope information is taken as label data to fine-tune model parameters in the trained amino acid sequence prediction model and the multilayer perceptron, to obtain a fine-tuned amino acid sequence prediction model and a fine-tuned multilayer perceptron. That is, the unoccluded double-chain sample biological information and the epitope information of the corresponding recognized sample cell receptor (the epitope information as label data) are taken as input data and input into the trained amino acid sequence prediction model and the multilayer perceptron, to further fine-tune the amino acid sequence prediction model and the multilayer perceptron that have been trained, so that the fine-tuned amino acid sequence prediction model and the fine-tuned multilayer perceptron can accurately extract features of the cell receptor and accurately recognize antigen specificity.

[0153] In the embodiments of the present application, by fine-tuning the model parameters in the trained amino acid sequence prediction model and the multilayer perceptron, on the one hand, the trained amino acid sequence prediction model can be made more suitable for the requirements of the antigen specificity recognition task, and on the other hand, the training of the multilayer perceptron can also be realized.

[0154] In the embodiments of the present application, after fine-tuning the model parameters in the trained amino acid sequence prediction model and the multilayer perceptron, a fine-tuned amino acid sequence prediction model and a fine-tuned multilayer perceptron are obtained. At this time, the fine-tuned amino acid sequence prediction model can be used to extract features of the cell receptor, and the fine-tuned multilayer perceptron can be used to determine the antigen specificity of the cell receptor. That is, the step of extracting features of the cell receptor based on the amino acid word sequence to obtain the amino acid sequence representation of the cell receptor can be realized by the fine-tuned amino acid sequence prediction model, and the step of determining the antigen specificity of the cell receptor based on the amino acid sequence representation can be realized by the fine-tuned multilayer perceptron.

[0155] The epitope information is explained as follows.

[0156] The epitope information is the real information of the sample cell receptor, and the epitope information is used as label information in the sample data for model fine-tuning. The epitope information includes the epitope of the sample cell receptor. The epitope is a chemical group existing on the surface of an antigen, which determines the specific structure of the antigen and is also called an antigenic determinant. The antigen binds to the corresponding lymphocyte surface antigen receptor through the epitope, thereby activating the lymphocyte and causing an immune response; the antigen also specifically binds to the corresponding antibody or sensitized lymphocyte. A single antigen molecule can have one or more different epitopes, and the size of the epitope is equivalent to the antigen binding site of the corresponding antibody, and each epitope has only one antigen specificity. Since the epitope in the epitope information is determined, the real antigen specificity of the sample cell receptor can be determined based on the epitope information.

[0157] Based on the training method of the amino acid sequence prediction model, the embodiment of the present application further provides a data preprocessing method for preprocessing the obtained pre-training data, Figure 6 is a flowchart of the implementation process of the data preprocessing method provided by the embodiment of the present application, as shown in Figure 6 The method comprises the following steps:

[0158] Step S601, from the pre-training data, data belonging to a specific object is screened out.

[0159] Here, the pre-training data is obtained from the database and comes from different objects. The target object to be screened out can be set as the specific object in advance. For example, if the human immune system is to be analyzed, the data of humans can be screened out.

[0160] Step S602, double-chain data pairing analysis is performed on the data belonging to the specific object to determine double-chain pairing data; the double-chain pairing data refers to data having two peptide chains that can be paired.

[0161] Here, the double-chain data pairing analysis refers to analyzing whether the data belonging to the specific object has two peptide chains that can be paired. When any data has two peptide chains that can be paired, the data is determined as double-chain pairing data; when any data does not have two peptide chains that can be paired, or only has single-chain data, the data is excluded from the data belonging to the specific object.

[0162] Step S603, data length analysis is performed on the double-chain pairing data to obtain the data length of each double-chain pairing data.

[0163] Here, the data length refers to the number of amino acids in the amino acid sequence corresponding to the double-chain paired data. For example, when the number of amino acids in the amino acid sequence (i.e., the double-chain paired data) corresponding to the two paired peptide chains of any particular object is 40, the data length of the double-chain paired data is 40.

[0164] In step S604, the double-chain paired data with a data length less than the length threshold is determined as the sample data.

[0165] Here, the double-chain paired data with a data length greater than or equal to the length threshold is removed to obtain the sample data. This is because, since the CDR3 sequence of the particular object has a certain length range, when the data length of any double-chain paired data exceeds the length range, it indicates that the double-chain paired data has data abnormalities or does not have the generality of model learning, and therefore the double-chain paired data can be removed.

[0166] For example, the length of the CDR3 region is usually less than 40, and therefore the CDR3 sequence with a length greater than 40 can be filtered out for the pre-training stage and the data of the downstream classification stage.

[0167] In some embodiments, the epitope information in the sample data can also be subjected to data removal processing. A threshold value of the epitope length can be set in advance, and the epitope length corresponding to the epitope information is determined. Here, the epitope length refers to the number of amino acids in the antigen epitope sequence corresponding to the epitope information. The epitope information with an epitope length greater than or equal to the epitope length threshold value can be deleted, i.e., the antigen epitope sequence with an epitope length greater than or equal to the epitope length threshold value is deleted.

[0168] In the embodiments of the present application, by sequentially performing particular object identification and screening, double-chain data pairing analysis and screening, length analysis and screening on the pre-training data obtained from the database, sample data capable of accurately training the amino acid sequence prediction model is obtained.

[0169] In the following, an exemplary application of the embodiments of the present application in an actual application scenario will be described.

[0170] The human immune system is composed of innate immunity and adaptive immunity. Adaptive immunity is a kind of immune response that can recognize and initiate against specific pathogens after contacting with the specific pathogens, and T cells and B cells are important components of the adaptive immune system. Antigen recognition is one of the key factors of T cell and B cell mediated immunity. T cells and B cells mainly interact with antigens by T cell receptors (TCR, a protein dimer) and B cell receptors (BCR) respectively. Among them, TCR recognizes and binds to the antigen presented by MHC on the cell membrane, while BCR directly binds to the specific antigen. TCR and BCR are both composed of two peptide chains (alpha and beta chains, or heavy and light chains), which form a loop three-dimensional structure of complementarity determining region (CDR) (including CDR1, CDR2 and CDR3) for antigen recognition and binding. When TCR recognizes MHC-antigen peptide complex, CDR1 and CDR2 recognize and bind to the side wall of the MHC molecule antigen binding groove, and CDR3 directly binds to the antigen peptide. Moreover, the CDR3 region has the largest variation, which directly determines the antigen binding specificity of TCR / BCR. Based on the sequence information and gene information of TCR / BCR, the antigen specificity is predicted, and based on the antigen specificity, the T cell / B cell activation ability is accurately predicted, which will have a revolutionary impact on the treatment of infectious diseases and autoimmune diseases, and the design of cancer immunization vaccine and many other frontier research fields. The difficulty of predicting the antigen specificity of TCR / BCR lies in: first, from the biological point of view, due to the random recombination of genes, there are many kinds of TCR / BCR in theory; second, from the perspective of machine learning training, there are less known antigen specificity data. Because there are less labeled data, and sequence data is used, BERT network of natural language processing can be used to learn "protein language" from supervision, so as to obtain the encoding of TCR / BCR, and then make downstream specificity prediction of TCR / BCR.

[0171] The embodiment of the present application provides a BERT-based immune receptor antigen specificity prediction method, which uses amino acid sequence information to predict the antigen specificity of TCR / BCR.

[0172] The antigen specificity prediction of TCR / BCR receptor is a core problem of immunology. Based on the sequence information and gene information of TCR / BCR, the antigen specificity is predicted, and based on the antigen specificity, the T cell / B cell activation ability is accurately predicted, which will have a revolutionary impact on the treatment of infectious diseases and autoimmune diseases, and the design of cancer immunization vaccine and many other frontier research fields.

[0173] Figure 7 The model framework diagram of the algorithm corresponding to the antigen specificity determination method provided by the embodiment of the present application is as follows, Figure 7As shown in the implementation process, the CDR3 region sequences of the two peptide chains are processed into kmer sequences respectively, each kmer represents a word, and the entire kmer sequence is regarded as a sentence. After the two sentences are connected, they are labeled, segmented, and position encoded (since BERT is a language model, a CDR3 sequence is naturally regarded as a sentence here, but there is no clear concept of a word in biological sequences, therefore, considering the length of the CDR3 sequence, three amino acids can be selected as a word; there are two amino acid overlaps between each two adjacent words).

[0174] Please continue to refer to Figure 7 Part A, the encoded TCR / BCR is input into a BERT model with 6 layers of data processing layers; this BERT model is pre-trained with larger data without epitope information (no label), and through this self-supervised technology, the protein language in the TCR / BCR is pre-learned into a model to extract some general features in the TCR / BCR sequence and obtain the encoding of the CDR3 sequence. It should be noted that for the BERT network, the following modifications can be made: since the pairing of the two peptide chains of TCR / BCR is relatively random, the NSP task of the BERT model is removed, and only the MLM task is retained. At the same time, the mask strategy is modified: since the kmer sequence is used as input data, there are two amino acid overlaps between words, so when masking a certain kmer, the left and right two words are also masked. As shown in Figure 8 , it is a schematic diagram of the masking effect before modifying the masking strategy. When performing masking processing, if the second word "AVP" is to be masked, and if only the second word "AVP" is masked, the model can easily infer this kmer through the left and right two words, for example, by inferring the second word "AVP" through "AV" in the first word "CAV" and "VP" in the third word "VPG". Therefore, when performing masking processing, when masking the second word "AVP", "AV" in the first word "CAV" also needs to be masked, and "VP" in the third word "VPG" also needs to be masked. Figure 9 , it is a schematic diagram of the masking effect after modifying the masking strategy. As can be seen, in order to prevent the model from easily inferring this kmer through the left and right two words, as shown in Figure 9 , the first word and the third word adjacent to the second word are both partially masked, which can avoid the model from converging too quickly during training. It should be noted that in the masking process as shown in Figure 9 , the last two amino acids of the first word are masked, and the first two amino acids of the third word are masked.

[0175] In the embodiments of the present application, the BERT model can be fine-tuned with data with epitope information, as shown inFigure 7 The B part shown, and through a Classification Head containing a multi-layer perceptron for mapping the learned multi-modal features to the final prediction of TCR / BCR antigen specificity.

[0176] The data preprocessing process of the algorithm corresponding to the antigen specificity determination method of the embodiments of the present application is described below.

[0177] For the pre-training data, such as the TCR / BCR pre-training data table shown in Table 1, the data of two databases, VDJDB and IEDB, and three data sets, 10xPBMC, nasopharyngeal carcinoma (NPC, Nasopharyngeal Carcinoma), and inflammatory bowel disease (IBD, Inflammatory Bowel Disease), were collected, and the data for the pre-training task were processed as follows: only human data were collected, and the data were required to be paired, with CDR3 sequences of alpha and beta chains; data with a length greater than 40 were also excluded, including TCR and BCR data. Because, on the one hand, there are some general rules that can be learned from the CDR3 sequences of TCR and BCR; on the other hand, it is hoped that the range of use of the pre-training model can be wider (can be applied to TCR and BCR tasks). For the data of the pre-training stage and the downstream classification stage, CDR3 sequences with a length greater than 40 (the length of the CDR3 region is usually less than 40) were screened out. For the data of the downstream classification stage, antigen epitope sequences with a length greater than 50 were also screened out. For the two data sets of the downstream classification task, the 10xPBMC data set came from 4 donations, and after being collected from the 10x website, quality control (QC) was performed, and data containing 7 epitopes were screened out, as shown in Table 2, downstream classification data-10xPBMC. The VDJ data set is data containing three epitopes collected from vdjdb; the data on vdjdb are usually data in published articles, which are reliable, as shown in Table 3, downstream classification data-VDJ. (It should be noted that 1049 / 5562 in Table 3 represents the number of TCRs combined with the epitope GILGFVFTL, which is 1049, and the total number of TCRs is 5562).

[0178] Table 1

[0179]

[0180]

[0181] Table 2

[0182]

[0183] Table 3

[0184] Epitope LTDEMIAQY TTDPSFLGRY YLQPRTFLL TCR data volume 132 / 592 254 / 592 206 / 592

[0185] After collecting pre-training data, the TCR / BCR sequence is processed into a kmer sequence. BERT is a language model; naturally, a CDR3 sequence can be regarded as a sentence, but there is no clear concept of a word in a biological sequence. In view of the foregoing and the length of the CDR3 sequence, 3 amino acids are selected as a word; there are two overlapping amino acids between each two adjacent words.

[0186] In the model training stage, the model (i.e., an amino acid sequence prediction model, a model provided based on BERT technology, denoted as sc-AIR-BERT) uses a cross-entropy function as a loss function for end-to-end training. In the task prediction stage, for the TCR / BCR antigen specificity recognition problem, the model can be directly predicted after training.

[0187] The embodiments of the present application compared the performance of the method in the related art on TCR antigen specificity recognition on the public data set, as shown in Figure 10 sc-AIR-BERT is the method proposed in the embodiments of the present application, and it can be seen that the AUC value of the prediction result of the sc-AIR-BERT model proposed in the embodiments of the present application is 0.835, which is greater than the AUC values of the prediction results of trc-BERT and DeepT RC models. Here, AUC refers to the possibility that the probability value of the classifier outputting a positive sample as positive is greater than the probability value of the classifier outputting a negative sample as positive, given a positive sample and a negative sample at random.

[0188] Figure 11 is a performance comparison diagram of the embodiments of the present application on TCR antigen specificity recognition on the public data set compared with the single chain method, as shown in Figure 11 α-sc-AIR-BERT is the performance of the method of the embodiments of the present application when only using α chain data, and β-sc-AIR-BERT is the performance of the method of the embodiments of the present application when only using β chain data. It can be seen that the AUC value of the prediction result of the model when using α chain and β chain data at the same time is greater than the performance when only using α chain data and the performance when only using β chain data.

[0189] Figure 12 is a performance comparison diagram of the embodiments of the present application on TCR antigen specificity recognition on the public data set compared with other encoding methods, as shown in Figure 12As shown, 2mer-BERT is the performance of the embodiment of the present application when using 2mer data, and 1mer-BERT is the performance of the embodiment of the present application when using 1mer data. As can be seen, the performance of using 3mer data (i.e., sc-AIR-BERT) corresponds to an AUC value greater than the performance of 2mer-sc-AIR-BERT and 1mer-sc-AIR-BERT, that is, the performance of using 3mer data is higher than the performance of using 2mer data and the performance of using 1mer data.

[0190] The antigen specificity determination method provided by the embodiment of the present application introduces the use of BERT pre-trained paired TCR / BCR sequences for adaptive immune receptor analysis, and uses the BERT network to learn the "protein language" in a self-supervised manner to obtain the encoding of the paired TCR / BCR sequence. The implementation of the embodiment of the present application provides technical support for the treatment of infectious diseases and autoimmune diseases, as well as many frontier research fields such as cancer immunovaccine design.

[0191] It should be noted that the pre-trained model of the embodiment of the present application can represent the amino acid sequence as encoding, and this pre-trained model applied to TCR / BCR specificity prediction can also be used for other downstream tasks, such as TCR / BCR affinity prediction, disease prediction, etc. The embodiment of the present application can also fuse V(D)J gene information for subsequent prediction tasks. In addition to kmer encoding, the encoding method used in data preprocessing in the embodiment of the present application can also use other encodings, including but not limited to Achely factor, kidera factor, one-hot encoding, etc.

[0192] It can be understood that in the embodiment of the present application, the content related to user information, such as double-stranded biological information, antigen specificity and disease-related information, etc., if it involves data related to user information or enterprise information, when the embodiment of the present application is applied to specific products or technologies, the user's permission or consent needs to be obtained, and the collection, use and processing of related data need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0193] The antigen specificity determination device 354 provided by the embodiment of the present application will be described below as an example of the exemplary structure of the software module, in some embodiments, as shown in Figure 2 As shown, the antigen specificity determination device 354 comprises:

[0194] The word encoding module is configured to perform word encoding processing on the double-chain biological information of the cell receptor to obtain an amino acid word sequence, wherein the amino acid word sequence includes at least one amino acid word representation; the feature extraction module is configured to perform feature extraction on the cell receptor based on the amino acid word sequence by using a pre-trained amino acid sequence prediction model; the amino acid sequence prediction model is obtained by training data obtained by performing occlusion processing on part of sample amino acid word representations in sample data; and the determination module is configured to determine the antigen specificity of the cell receptor based on the amino acid word sequence.

[0195] In some embodiments, the word encoding module is further configured to: obtain two peptide chains of the cell receptor, and determine the double-chain biological information of the cell receptor based on the two peptide chains; determine a monomer unit quantity corresponding to the word encoding processing; encode consecutive amino acids of the double-chain biological information of the cell receptor per monomer unit quantity into an amino acid word representation to form the amino acid word sequence; and wherein each two adjacent amino acid word representations have a preset number of overlapping amino acids, and the preset number is less than the monomer unit quantity by 1.

[0196] In some embodiments, the cell receptor includes a T cell receptor or a B cell receptor; when the cell receptor is a T cell receptor, the two peptide chains include an alpha chain and a beta chain; when the cell receptor is a B cell receptor, the two peptide chains include a heavy chain and a light chain; and the monomer unit quantity is 3.

[0197] In some embodiments, the double-chain biological information includes amino acid information corresponding to each of the two peptide chains; and the word encoding module is further configured to: encode consecutive amino acids of each of the two peptide chains per monomer unit quantity into an amino acid word representation to form an amino acid subsequence corresponding to each of the two peptide chains; and determine the amino acid word sequence based on the amino acid subsequences corresponding to the two peptide chains.

[0198] In some embodiments, the word encoding module is further configured to: perform splicing processing on the amino acid subsequences corresponding to the two peptide chains to obtain a spliced word sequence; and perform, in sequence, marking processing, segmentation processing, and position encoding processing on the spliced word sequence to obtain the amino acid word sequence.

[0199] In some embodiments, the feature extraction module is further configured to perform feature extraction on the cell receptor based on the amino acid word sequence by using the pre-trained amino acid sequence prediction model to obtain an amino acid sequence representation of the cell receptor.

[0200] In some embodiments, the apparatus further comprises a model training module configured to train the amino acid sequence prediction model; wherein the model training module is further configured to: perform data preprocessing on the obtained pre-training data to obtain sample data; wherein the sample data comprises sample double-chain biological information of a sample cell receptor; input the sample double-chain biological information into the amino acid sequence prediction model; perform word encoding processing on the sample double-chain biological information by a word encoding processing layer of the amino acid sequence prediction model to obtain a sample amino acid word sequence; wherein the sample amino acid word sequence comprises at least one sample amino acid word representation; perform occlusion processing on the at least one sample amino acid word representation in the sample amino acid word sequence by an occlusion processing layer of the amino acid sequence prediction model to obtain an occluded sample amino acid word sequence; perform amino acid sequence prediction on the sample cell receptor based on the occluded sample amino acid word sequence by a prediction processing layer of the amino acid sequence prediction model to determine the sample amino acid word representation occluded during the occlusion processing, and obtain a sample amino acid sequence representation of the sample cell receptor; input the sample amino acid sequence representation into a preset loss model to obtain a loss result; and correct model parameters in the word encoding processing layer, the occlusion processing layer, and the prediction processing layer based on the loss result to obtain a trained amino acid sequence prediction model.

[0201] In some embodiments, determining the antigen specificity of the cell receptor is achieved by a multilayer perceptron; the model training module is further configured to: obtain fine-tuning sample data comprising unoccluded double-chain sample biological information and epitope information of a sample cell receptor; use the epitope information as label data to fine-tune model parameters in the trained amino acid sequence prediction model and the multilayer perceptron to obtain a fine-tuned amino acid sequence prediction model and a fine-tuned multilayer perceptron; accordingly, the feature extraction is further configured to extract features of the cell receptor using the fine-tuned amino acid sequence prediction model, and the determination module is further configured to determine the antigen specificity of the cell receptor using the fine-tuned multilayer perceptron.

[0202] In some embodiments, the model training module is further configured to: determine, from the sample amino acid word sequence, a randomly selected at least one sample amino acid word representation as a target amino acid word representation; determine adjacent amino acid word representations adjacent to the target amino acid word representation; wherein the adjacent amino acid word representations include a first adjacent amino acid word representation adjacent to a first side of the target amino acid word representation, and a second adjacent amino acid word representation adjacent to a second side of the target amino acid word representation; when the target amino acid word representation is located at a sequence start position of the amino acid word sequence, the adjacent amino acid word representations include a second adjacent amino acid word representation adjacent to a second side of the target amino acid word representation; when the target amino acid word representation is located at a sequence end position of the amino acid word sequence, the adjacent amino acid word representations include a first adjacent amino acid word representation adjacent to a first side of the target amino acid word representation; and perform occlusion processing on the target amino acid word representation, overlapping amino acids in the first adjacent amino acid word representation, and overlapping amino acids in the second adjacent amino acid word representation, to obtain an occluded sample amino acid word sequence.

[0203] In some embodiments, the model training module is further configured to: input the sample amino acid sequence representation and the sample double-chain biological information into the preset loss model; determine, by a cross-entropy loss function in the preset loss model, a sequence distance between the sample amino acid sequence representation and the sample double-chain biological information; and determine, according to the sequence distance, the loss result.

[0204] In some embodiments, the amino acid sequence representation is a multi-modal feature; and the determination module is further configured to: input the amino acid sequence into the multi-layer perceptron; and perform mapping processing on the multi-modal feature corresponding to the amino acid sequence by the multi-layer perceptron, to obtain the antigen specificity of the cell receptor.

[0205] In some embodiments, the model training module is further configured to: filter, from the pre-training data, data belonging to a specific object; perform double-chain data pairing analysis on the data belonging to the specific object, to determine double-chain pairing data; the double-chain pairing data refers to data of two peptide chains having a pairing; perform data length analysis on the double-chain pairing data, to obtain a data length of each double-chain pairing data; and determine, as the sample data, double-chain pairing data having a data length less than a length threshold.

[0206] It should be noted that the description of the device embodiments of the present application is similar to the description of the above method embodiments, and has similar beneficial effects as the method embodiments, and thus is not described herein. For technical details not disclosed in the device embodiments, please refer to the description of the method embodiments for understanding.

[0207] The embodiment of the present application provides a computer program product or computer program, the computer program product or computer program comprises executable instructions, the executable instructions are computer instructions; the executable instructions are stored in a computer readable storage medium. When the processor of the antigen specificity determination device reads the executable instructions from the computer readable storage medium, the processor executes the executable instructions, so that the antigen specificity determination device executes the method provided in the embodiment of the present application.

[0208] The embodiment of the present application provides a storage medium storing executable instructions, wherein the executable instructions are stored, and when the executable instructions are executed by a processor, the processor will execute the method provided by the embodiment of the present application, for example, the method shown in the embodiment of the present application. Figure 3

[0209] In some embodiments, the storage medium can be a computer readable storage medium, for example, ferroelectric memory (FRAM, Ferromagnetic Random Access Memory), read-only memory (ROM, Read Only Memory), programmable read-only memory (PROM, Programmable Read Only Memory), erasable programmable read-only memory (E PROM, Erasable Programmable Read Only Memory), electrically erasable programmable read-only memory (EEPR OM, Electrically Erasable Programmable Read Only Memory), flash memory, magnetic surface memory, optical disc, or compact disc read-only memory (CD-ROM, Compact Disk-Read Only Memory) and the like memory; it can also be various devices comprising one or any combination of the above memories.

[0210] In some embodiments, the executable instructions can be in the form of programs, software, software modules, scripts or codes, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and can be deployed in any form, including being deployed as independent programs or being deployed as modules, components, subroutines or other units suitable for use in a computing environment.

[0211] ​As an example, executable instructions may, but do not necessarily, correspond to files in a file system. They may be stored as part of a file containing other programs or data, for example, in one or more scripts within a Hypertext Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., files storing one or more modules, subroutines, or code sections). As an example, executable instructions may be deployed to execute on a single electronic device, or on multiple electronic devices located in one location, or on multiple electronic devices distributed across multiple locations and interconnected via a communication network.

[0212] The above description is merely an embodiment of this application and is not intended to limit the scope of protection of this application. Any modifications, equivalent substitutions, and improvements made within the spirit and scope of this application are included within the scope of protection of this application.

Claims

1. A method for determining antigen specificity, characterized in that, The method includes: Two peptide chains of a cell receptor are obtained, and the double-stranded biological information of the cell receptor is determined based on the two peptide chains. Determine the number of single-unit words corresponding to word encoding processing; The number of consecutive amino acids per monomer unit in the double-stranded biological information of the cell receptor is encoded into an amino acid word representation, forming an amino acid word sequence; wherein, the amino acid word sequence includes at least one amino acid word representation; there is a predetermined number of overlapping amino acids between every two adjacent amino acid word representations, the predetermined number being 1 less than the number of monomer units; A pre-trained amino acid sequence prediction model is used to extract features from the cell receptor based on the amino acid word sequence to obtain the amino acid sequence representation of the cell receptor. The amino acid sequence prediction model is trained on data obtained after masking a portion of the sample amino acid word representations. The masking process involves covering up information from the amino acid word representations of samples before and after the current sample amino acid word representation. The antigen specificity of the cell receptor is determined based on the amino acid sequence representation.

2. The method according to claim 1, characterized in that, The cell receptor includes a T cell receptor or a B cell receptor; when the cell receptor is a T cell receptor, the two peptide chains include an α chain and a β chain; when the cell receptor is a B cell receptor, the two peptide chains include a heavy chain and a light chain. The number of individual units is 3.

3. The method according to claim 1, characterized in that, The double-stranded biological information includes the amino acid information corresponding to each of the two peptide chains; The step of encoding a number of consecutive amino acids per monomer unit in the double-stranded biological information of the cell receptor into an amino acid word to form the amino acid word sequence includes: Each number of consecutive amino acids in each monomer unit of each peptide chain is encoded into an amino acid word, which forms the corresponding amino acid subsequence for each peptide chain. The amino acid sequence is determined based on the amino acid subsequences corresponding to the two peptide chains.

4. The method according to claim 3, characterized in that, The step of determining the amino acid sequence based on the amino acid sub-sequences corresponding to the two peptide chains includes: The amino acid subsequences corresponding to the two peptide chains are spliced ​​together to obtain the splicing word sequence; The concatenated word sequence is sequentially labeled, segmented, and positionally encoded to obtain the amino acid word sequence.

5. The method according to any one of claims 1 to 4, characterized in that, The step of extracting features from the cell receptor based on the amino acid sequence to obtain the amino acid sequence representation of the cell receptor includes: The cell receptor is characterized by feature extraction based on the amino acid word sequence using a pre-trained amino acid sequence prediction model, thereby obtaining the amino acid sequence representation of the cell receptor.

6. The method according to claim 5, characterized in that, The amino acid sequence prediction model is trained in the following way: The acquired pre-training data is preprocessed to obtain sample data; wherein, the sample data includes sample double-stranded biological information of sample cell receptors; The biological information of the sample double strand is input into the amino acid sequence prediction model; The word encoding processing layer of the amino acid sequence prediction model is used to perform word encoding processing on the sample double-stranded biological information to obtain the sample amino acid word sequence; wherein, the sample amino acid word sequence includes at least one sample amino acid word representation; The masking processing layer of the amino acid sequence prediction model is used to mask at least one sample amino acid word representation in the sample amino acid word sequence to obtain the masked sample amino acid word sequence. The amino acid sequence prediction model uses a prediction processing layer to predict the amino acid sequence of the sample cell receptor based on the masked sample amino acid word sequence, in order to determine the masked sample amino acid word representation during the masking process, and thus obtain the sample amino acid sequence representation of the sample cell receptor. The sample amino acid sequence representation is input into a preset loss model to obtain the loss result; Based on the loss result, the model parameters in the word encoding processing layer, the occlusion processing layer and the prediction processing layer are corrected to obtain the trained amino acid sequence prediction model.

7. The method according to claim 6, characterized in that, The determination of the antigen specificity of the cell receptor is achieved by a multilayer sensor; the method further includes: Acquire fine-tuned sample data, which includes unobstructed double-stranded sample biological information and epitope information of sample cell receptors; Using the epitope information as label data, the model parameters in the trained amino acid sequence prediction model and the multilayer perceptron are fine-tuned to obtain the fine-tuned amino acid sequence prediction model and the fine-tuned multilayer perceptron. Accordingly, the cell receptor is characterized by the fine-tuned amino acid sequence prediction model, and the antigen specificity of the cell receptor is determined by the fine-tuned multilayer perceptron.

8. The method according to claim 6, characterized in that, The process of masking at least one sample amino acid word representation in the sample amino acid word sequence to obtain a masked sample amino acid word sequence includes: From the sample amino acid word sequence, at least one randomly selected sample amino acid word representation is determined as the target amino acid word representation; Determine adjacent amino acid word representations adjacent to the target amino acid word representation; wherein, the adjacent amino acid word representations include: a first adjacent amino acid word representation adjacent to a first side of the target amino acid word representation, and a second adjacent amino acid word representation adjacent to a second side of the target amino acid word representation; when the target amino acid word representation is located at the beginning position of the amino acid word sequence, the adjacent amino acid word representations include the second adjacent amino acid word representation adjacent to the second side of the target amino acid word representation; when the target amino acid word representation is located at the end position of the amino acid word sequence, the adjacent amino acid word representations include the first adjacent amino acid word representation adjacent to the first side of the target amino acid word representation; The overlapping amino acids in the target amino acid word representation, the first adjacent amino acid word representation, and the second adjacent amino acid word representation are masked to obtain the masked sample amino acid word sequence.

9. The method according to claim 6, characterized in that, The step of inputting the sample amino acid sequence representation and the epitope information into a preset loss model to obtain the loss result includes: The amino acid sequence representation of the sample and the double-stranded biological information of the sample are input into the preset loss model; The sequence distance between the sample amino acid sequence representation and the sample double-stranded biological information is determined by the cross-entropy loss function in the preset loss model. The loss result is determined based on the sequence distance.

10. The method according to claim 7, characterized in that, The amino acid sequence is represented as a multimodal feature; the determination of the antigen specificity of the cell receptor based on the amino acid sequence representation includes: The amino acid sequence is input into the multilayer sensor; The antigen specificity of the cell receptor is obtained by mapping the multimodal features corresponding to the amino acid sequence through the multilayer sensor.

11. The method according to claim 6, characterized in that, The process of preprocessing the acquired pre-training data to obtain sample data includes: From the pre-trained data, data belonging to a specific object is selected; Double-strand data pairing analysis is performed on the data belonging to a specific object to determine the double-strand pairing data; the double-strand pairing data refers to data with two paired peptide chains; Data length analysis is performed on the double-stranded pairing data to obtain the data length of each double-stranded pairing data; The double-chain paired data with a length less than the length threshold are identified as the sample data.

12. An antigen-specific determination device, characterized in that, The device includes: A word encoding module is used to acquire two peptide chains of a cell receptor and determine the double-stranded biological information of the cell receptor based on the two peptide chains; determine the number of monomer units corresponding to the word encoding processing; encode the number of consecutive amino acids of each monomer unit in the double-stranded biological information of the cell receptor into an amino acid word representation, forming an amino acid word sequence; wherein, the amino acid word sequence includes at least one amino acid word representation; there is a preset number of overlapping amino acids between every two adjacent amino acid word representations, the preset number being 1 less than the number of monomer units; The feature extraction module is used to extract features from the cell receptor based on the amino acid word sequence using a pre-trained amino acid sequence prediction model to obtain the amino acid sequence representation of the cell receptor; wherein, the amino acid sequence prediction model is obtained by training on data after masking a portion of the sample amino acid word representations in the sample data; the masking process involves masking the information of the sample amino acid word representations at positions before and after the current sample amino acid word representation; A determination module is used to determine the antigen specificity of the cell receptor based on the amino acid sequence representation.

13. The apparatus according to claim 12, characterized in that, The cell receptor includes a T-cell receptor or a B-cell receptor; when the cell receptor is a T-cell receptor, the two peptide chains include an α-chain and a β-chain; when the cell receptor is a B-cell receptor, the two peptide chains include a heavy chain and a light chain; the number of monomer units is 3.

14. The apparatus according to claim 12, characterized in that, The double-stranded biological information includes the amino acid information corresponding to each of the two peptide chains; the word encoding module is further used to: encode the number of consecutive amino acids per monomer unit in each peptide chain into an amino acid word representation, thereby forming an amino acid subsequence corresponding to each peptide chain; and determine the amino acid word sequence based on the amino acid subsequences corresponding to the two peptide chains.

15. The apparatus according to claim 12, characterized in that, The word encoding module is further used to: splice the amino acid subsequences corresponding to the two peptide chains to obtain the spliced ​​word sequence; and sequentially perform labeling, segmentation and position encoding on the spliced ​​word sequence to obtain the amino acid word sequence.

16. An antigen-specific determination device, characterized in that, include: Memory, used to store executable instructions; A processor, when executing executable instructions stored in the memory, implements the antigen-specific determination method according to any one of claims 1 to 11.

17. A computer-readable storage medium, characterized in that, The device stores executable instructions for causing a processor to execute the executable instructions to implement the antigen-specific determination method according to any one of claims 1 to 11.

18. A computer program product, characterized in that, The computer program product includes executable instructions stored in a computer-readable storage medium; wherein, when the processor of the antigen-specific determination device reads the executable instructions from the computer-readable storage medium and executes the executable instructions, it implements the antigen-specific determination method according to any one of claims 1 to 11.

Images