Method and system for predicting affinity of antibody mutant and antigen

Abstract

The invention discloses a method and system for predicting affinity of an antibody mutant and an antigen, and the method comprises the following steps: (1) collecting affinity experimental data of an antibody and interaction of the antibody and the antigen, and constructing a fine tuning data set; (2) training a UniRep model by using the fine tuning data set to obtain an affinity prediction model of the antibody and the antigen; (3) mutating at key sites of the antibody sequence to obtain an antibody mutant; and predicting the affinity between the antibody mutant and the antigen by using the affinity prediction model. According to the method and the system, the change of antigen affinity of the antibody after mutation at different sites can be accurately evaluated, the experiment screening cost is reduced, and the antibody design efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics and protein engineering technology, and in particular to a method and system for predicting the affinity between antibody mutants and antigens by means of a large protein language model. Background Art

[0002] The affinity between antibodies and antigens is a key factor in determining the effectiveness of immune responses. Studying the effects of antibody mutations on their affinity for specific antigens is crucial for antibody drug development and vaccine design. However, due to the complexity of antibody-antigen interactions and the vastness of the antibody mutation space, experimental screening of antibody mutants is costly and inefficient. Therefore, it is particularly important to develop a computational model that can effectively predict changes in antibody affinity with antigens after mutation.

[0003] Currently, there are multiple models for predicting the affinity between antibodies and antigens. For example, the DeepAb model extracts antibody sequence features through convolutional neural networks (CNNs) and combines the structural information of antibodies and antigens for affinity prediction; the TITAN model uses graph neural networks (GNNs) to combine the three-dimensional structural data of the antibody and antigen interface to model their interactions. However, these models have some limitations in application, especially that most of them rely on the three-dimensional structural data of antibodies or antigens. This leads to great restrictions on the application of these models in the absence of structural data. In addition, when dealing with complex multi-site mutations, the prediction accuracy of these models tends to decrease, and their computational cost is high, making it difficult to meet the needs of large-scale mutation screening. Relevant prior arts include: for example, a Chinese patent document with publication number CN 117334247 A discloses a training method for an antigen-antibody affinity prediction model and an antibody screening method, which involves artificial intelligence fields such as natural language processing and deep learning, including obtaining a target complex three-dimensional feature map extraction model and a first model parameter set of the target complex three-dimensional feature map extraction model; assigning parameters to the initial feature map extraction layer based on the first model parameter set to obtain a candidate affinity prediction model to be trained; obtaining a first training sample of the candidate affinity prediction model including a first complex sample and a mutant complex sample set of the first complex sample; inputting the first complex sample and the mutant complex sample set into the candidate affinity prediction model for model training until the training is completed to obtain a trained target affinity prediction model.

[0004] In order to solve these problems, protein language models came into being. The development of this type of model began in 2019. Inspired by language models in natural language processing (such as Word2Vec and BERT), protein sequences were first regarded as "languages" and amino acids as "words". The structural and functional information of sequences was extracted from a large number of protein sequences through deep learning. Representative protein language models include UniRep, ESM (Evolutionary Scale Modeling) and ProtTrans, which can capture complex patterns in protein sequences through large-scale pre-training and are applied to prediction tasks such as folding, function, and activity.

[0005] A significant advantage of protein language models is that they can extract useful features from protein sequences without relying on three-dimensional structural data. This allows efficient and accurate affinity prediction in the absence of structural data. In addition, such models show strong representation capabilities when dealing with complex sequence changes such as multi-site mutations, and can quickly handle large-scale mutation screening tasks. Through these features, protein language models overcome the limitations of existing models and significantly improve computational efficiency and prediction accuracy. Summary of the invention

[0006] The present invention provides a method and system for predicting the affinity of antibody mutants to antigens, which can accurately evaluate the changes in the affinity of antibodies to antigens after mutations at different sites, reduce experimental screening costs, and improve antibody design efficiency.

[0007] The technical solution of the present invention is as follows: A method for predicting the affinity of an antibody mutant to an antigen comprises the following steps: (1) Collect affinity experimental data of antibodies and antibody-antigen interactions and construct a fine-tuning dataset; (2) Use the fine-tuning dataset to train the UniRep model to obtain the affinity prediction model between antibodies and antigens; (3) Mutate key sites in the antibody sequence to obtain antibody mutants; use the affinity prediction model to predict the affinity between the antibody mutants and the antigen.

[0008] The fine-tuning data set includes the amino acid sequence of the antibody and the affinity measurement value of the interaction between the antibody and the antigen.

[0009] Each antibody sequence is represented by its amino acid sequence, which is represented by single-letter symbols (such as A, C, D, E, etc.). Each amino acid in the antibody sequence is used as input to the UniRep model.

[0010] Step (2) includes: (2-1) Divide the fine-tuning dataset into training set and test set; (2-2) The training set is used to supervise the learning of the UniRep model, and the UniRep model parameters are gradually optimized by minimizing the error between the affinity prediction value and the actual affinity measurement value; after the training is completed, the affinity prediction model is obtained; (2-3) The test set is used to evaluate the model’s predictive ability on unseen data.

[0011] In step (2-1), the fine-tuning dataset is divided into K subsets of equal size using the K-fold cross-validation strategy. One subset is used as the test set each time, and the remaining K-1 subsets are used as training sets.

[0012] Step (2-2) includes: (2-2i) Antibody X i The amino acid sequence of antibody X was input into the UniRep model, and the UniRep model extracted i The sequence features of antibody X are predicted based on the sequence features. i Affinity of interaction with antigen; formula: ;

[0013] in, is the ith antibody X i UniRep(X) represents the predicted affinity value of antibody X by UniRep model. i The amino acid sequence is extracted from the sequence features; f is the regression function, which represents the mapping relationship from sequence features to affinity; ϵ is the prediction error term; (2-2ii) The mean square error (MSE) was used as the loss function to calculate the error between the predicted affinity value and the measured affinity value: ;

[0014] Among them, Y i is the actual affinity value measured in the experiment, is the predicted value of the model, and N is the number of samples in the test set; (2-2iii) During the model training process, the gradient of the loss function to the model parameters is calculated through the back propagation algorithm, and the model parameters are updated using the gradient descent algorithm; the specific optimization formula is as follows: ;

[0015] in, i t is the current model parameter; or is the learning rate; is the gradient of the loss function with respect to the model parameters; (2-2iv) Through continuous iteration and optimization, an affinity prediction model is obtained.

[0016] MSE, as a loss function, can effectively measure the difference between the model prediction value and the real experimental data. By minimizing MSE, the model is able to adjust parameters to improve the accuracy of affinity prediction.

[0017] In step (2-2i), the UniRep model extracts antibody X i The sequence features include: i The amino acid sequence is converted into an amino acid feature vector; The amino acid features of each time step are processed step by step through the LSTM network to generate the context-dependent features of the amino acid sequence; the formula is expressed as: ;

[0018] Among them, h t represents the hidden state of the LSTM network at time step t; x t represents the input features of the antibody amino acid sequence at time step t; h t-1 represents the hidden state of the previous time step t-1; The hidden state of the last time step is the context-dependent feature of the amino acid sequence, which is antibody X i sequence characteristics.

[0019] In step (3), the mutation at the key site of the antibody sequence includes: obtaining antibody mutants by single point mutation or multiple point mutation in the complementarity determining region of the antibody sequence.

[0020] In step (3), the affinity prediction model is used to predict the affinity between the antibody mutant and the antigen, including: inputting the amino acid sequence of the antibody mutant into the affinity prediction model, the affinity prediction model extracts the sequence features of the antibody mutant, and predicts the affinity of the interaction between the antibody mutant and the antigen based on the sequence features, the formula is: ;

[0021] in, Is the affinity prediction model for antibody mutant X mut The affinity prediction value of UniRep(X mut ) represents the sequence features of the antibody mutants extracted by the affinity prediction model; f is a regression function that directly maps the extracted features to affinity values.

[0022] The antibody is an anti-HER2 antibody; the antigen is a HER2 antigen.

[0023] The present invention also provides a system for predicting the affinity of antibody mutants to antigens, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory also stores the affinity prediction model, and when the processor executes the computer program, the following steps are implemented: The amino acid sequence of the antibody mutant is input into the affinity prediction model, and the predicted value of the affinity between the antibody mutant and the antigen is output.

[0024] Compared with the prior art, the present invention has the following beneficial effects: Based on the protein language model UniRep model, the present invention proposes an innovative method and system for predicting antibody mutants and antigen affinity. Unlike traditional models that rely on three-dimensional structural information, the present invention only relies on the sequence data of antibodies and antigens, can flexibly handle complex mutation situations, and adapt to the prediction tasks of specific antigens (such as HER2) through fine-tuning after large-scale training.

[0025] The present invention not only improves the affinity prediction accuracy of antibody mutants, but also significantly accelerates large-scale mutation screening tasks, reduces experimental costs and time in antibody drug development, and has broad application prospects and technical advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic diagram of the overall framework principle of the present invention. DETAILED DESCRIPTION

[0027] The embodiments of the present invention are described in detail below, wherein the same or similar reference numerals throughout represent the same or similar elements or elements with similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are only used to explain the present invention and cannot be used as a limitation of the present invention.

[0028] It will be understood by those skilled in the art that, unless otherwise defined, all terms (including technical terms and scientific terms) used herein have the same meaning as generally understood by those skilled in the art in the field to which the present invention belongs. It should also be understood that terms such as those defined in common dictionaries should be understood to have meanings consistent with the meanings in the context of the prior art, and will not be interpreted with idealized or overly formal meanings unless defined as herein.

[0029] The numbers of the steps mentioned in each embodiment are only for the convenience of description and have no substantial sequence relationship. Different steps in each specific embodiment can be combined in different sequences to achieve the purpose of the invention.

[0030] The present invention is further described below in conjunction with the accompanying drawings and specific embodiments.

[0031] Example: like Figure 1 As shown, a method for predicting the affinity between antibody mutants and antigens by using a protein language large model specifically comprises the following steps: 101) Steps to construct an antibody affinity prediction model First, the long short-term memory network (LSTM) in the UniRep model is used to extract the deep features of the antibody sequence and build an affinity prediction model between the antibody and the antigen. UniRep captures the long-range dependency information by processing each amino acid site in the antibody sequence, thereby obtaining the key features of the interaction between the antibody sequence and the antigen. The antibody sequence data is the input raw data, and the LSTM network is used to learn the characteristics of the antibody sequence and gradually generate local and global feature representations of the sequence through multi-layer recursion. These features can reflect the structure of the antibody and the potential impact of its functional regions, thereby laying the foundation for antibody affinity prediction.

[0032] Furthermore, the processing flow of the LSTM network is as follows: Input antibody sequence: Each antibody sequence is represented by its amino acid sequence, which is represented by single-letter symbols (such as A, C, D, E, etc.), and each amino acid in the antibody sequence is used as input. The UniRep model can handle variable-length protein sequences, so there is no need to perform special cropping or padding on the sequence length, and then convert it into a feature vector of a specific dimension through the embedding layer. Each amino acid sequence is represented as a continuous vector after embedding, which captures the semantic relationship between amino acids.

[0033] LSTM processing features: The LSTM network generates a hidden state for each time step by processing the amino acid features of each time step in the sequence. The hidden state is the core of LSTM to capture contextual information. By combining the information of the previous time step (hidden state), the current state is gradually updated. The hidden state not only reflects the information of the current amino acid, but also incorporates the influence of the previous amino acid, thereby capturing the long-range dependency of the antibody sequence.

[0034] Global and local feature output: LSTM finally outputs the hidden state representation of the sequence, which contains both local feature representation (i.e., sequence information of the current time step) and global feature representation (i.e., contextual dependency of the entire sequence). The UniRep model represents global features through the hidden state at the last moment, and reflects the overall features of the sequence through the average hidden state. These features will be used in the subsequent affinity prediction model.

[0035] ; Among them, h t represents the hidden state of LSTM at time step t, x trepresents the input features of the antibody sequence at time step t, h t-1 Represents the hidden state at the previous moment.

[0036] Each hidden state h t The calculation will combine the input feature x of the current amino acid t and the hidden state h at the previous moment t-1 , which recursively processes antibody sequences to generate deep features representing antibody functional regions.

[0037] 102) Dataset preparation and fine-tuning First, a dataset containing experimental data on the affinity of anti-HER2 antibodies to HER2 antigens was selected. This dataset includes antibody sequences and their affinity measurements for their interaction with HER2 antigens. To ensure that the model can be effectively generalized in various situations, the dataset needs to be divided into a training set and a test set. The training set is used for supervised learning of the UniRep model, enabling it to learn key features from the antibody sequences, while the test set is used to evaluate the model's predictive ability on unseen data.

[0038] Specifically, the training set is used to minimize the error between the predicted affinity value and the experimental true value. By optimizing the model parameters, it can effectively capture the structural characteristics of the antibody sequence and its binding relationship with the HER2 antigen. In this process, the UniRep model uses its embedding layer and LSTM network to feature each amino acid in the antibody sequence, thereby generating a deep feature representation of the sequence.

[0039] During the model fine-tuning process, the UniRep model first encodes each amino acid in the antibody sequence into a high-dimensional feature vector through the embedding layer, and then uses the LSTM network to process these feature vectors to generate context-dependent features of the sequence. Finally, these features are mapped into affinity prediction values ​​through the fully connected layer. The prediction formula is: ; in is the i-th antibody sequence X i The predicted affinity value of UniRep(X) represents the extraction of sequence features through the UniRep model; f is the regression function, which represents the mapping relationship from features to affinity; ϵ is the prediction error term.

[0040] At this stage, the core task of the model is to minimize the gap between the predicted value and the actual experimental data by optimizing the model parameters.

[0041] Loss function and optimization,In order to optimize the parameters of the UniRep model, the mean square error (MSE) is used as the loss function.,MSE can quantify the difference between the model's predicted value and the true value,,helping the model gradually adjust the parameters to improve the,accuracy of the prediction. The specific formula of the loss function is: ; Among them, Y i is the actual affinity value measured in the experiment, is the predicted value of the model, and N is the number of samples. MSE helps the model continuously adjust its internal weights by measuring the error between the predicted value and the true value, thereby improving the prediction performance. During the model training process, the gradient of the loss function to the model parameters is calculated through the back propagation algorithm, and the model parameters are updated using the gradient descent algorithm. The specific optimization formula is as follows: ; in i t is the current model parameter; or is the learning rate, which is used to control the step size of parameter update; is the gradient of the loss function with respect to the model parameters. Through continuous iteration and optimization, the model can learn the deep features related to affinity in the antibody sequence and gradually improve the performance of the model in the affinity prediction task.

[0042] Dataset partitioning and cross-validation,In order to ensure the generalization ability of the model, the K-fold cross-validation strategy is adopted when partitioning the data set. K-fold cross-validation divides the data set into K subsets of equal size, and uses one subset as the test set each time, and the remaining K-1 subsets as the training set. Through cross-validation, the model can be trained and tested under different data partitions, further verifying its robustness and generalization ability on different data.

[0043] 103) Mutant generation and affinity prediction After the model has been fine-tuned and trained, the next step is to generate antibody mutants and predict affinity. Mutation design is performed at key sites in the antibody sequence (such as the complementary determining region, CDR). The CDR region is critical for the binding of antibodies to antigens, and mutations may have a significant effect on the binding ability of antibodies. Therefore, mutation design is usually performed at key sites in the CDR region, which may simulate different antibody variants through single-point mutations or multi-point mutations.

[0044] After the mutation design is completed, the mutant sequence is input into the fine-tuned UniRep model. The model predicts the changes in affinity between the mutant sequences and the HER2 antigen based on the characteristics of the mutant sequences.

[0045] ; in, is the fine-tuned UniRep model for the mutant antibody sequence X mut The affinity prediction value of UniRep(X mut ) represents the feature representation of the mutant sequence extracted by the UniRep model, and f is a regression function that directly maps the extracted features to affinity values.

[0046] In this step, the model does not perform further error calculations or model optimization, and the fine-tuned UniRep model directly outputs the predicted affinity value of the mutant. This prediction process is very efficient and can handle large-scale mutant data, helping researchers quickly screen out antibody mutants with excellent performance.

[0047] Mutant screening and application: By predicting the changes in affinity between different mutants and HER2 antigens, antibody mutants with high affinity can be quickly screened out. The screened mutants can be used for further experimental verification or applied in the early stages of antibody drug development. Through the screening and optimization of mutants, researchers can more efficiently design antibody drugs with better therapeutic potential.

[0048] In summary, step 103) generates affinity prediction results of antibody mutants based on the fine-tuned UniRep model, and obtains high-affinity antibody mutants through rapid screening, which provides important technical support for the development and optimization of antibody drugs.

[0049] Although the above methods are illustrated and described as a series of actions for simplicity of explanation, it should be understood and appreciated that these methods are not limited by the order of the actions, because according to one or more embodiments, some actions may occur in a different order and / or concurrently with other actions from those illustrated and described herein or not illustrated and described herein but understandable to those skilled in the art.

[0050] The present invention also provides a system for predicting the affinity of antibody mutants to antigens, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the memory also stores the affinity prediction model, and when the processor executes the computer program, the following steps are implemented: The amino acid sequence of the antibody mutant is input into the affinity prediction model, and the predicted value of the affinity between the antibody mutant and the antigen is output.

[0051] Those skilled in the art will further appreciate that the various illustrative logic blocks, modules, circuits, and algorithm steps described in conjunction with the embodiments disclosed herein may be implemented as electronic hardware, computer software, or a combination of the two. To clearly illustrate this interchangeability of hardware and software, various illustrative components, blocks, modules, circuits, and steps are generally described above in terms of their functionality. Whether such functionality is implemented as hardware or software depends on the specific application and the design constraints imposed on the overall system. The technician may implement the described functionality in different ways for each specific application, but such implementation decisions should not be interpreted as resulting in a departure from the scope of the present invention.

[0052] The various illustrative logic blocks, modules, and circuits described in conjunction with the embodiments disclosed herein may be implemented or executed with a general purpose processor, a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic device, discrete gate or transistor logic, discrete hardware components, or any combination thereof designed to perform the functions described herein. A general purpose processor may be a microprocessor, but in an alternative, the processor may be any conventional processor, controller, microcontroller, or state machine. The processor may also be implemented as a combination of computing devices, such as a combination of a DSP and a microprocessor, a plurality of microprocessors, one or more microprocessors in cooperation with a DSP core, or any other such configuration.

[0053] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be embodied directly in hardware, in a software module executed by a processor, or in a combination of the two. The software module may reside in a RAM memory, a flash memory, a ROM memory, an EPROM memory, an EEPROM memory, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art. An exemplary storage medium is coupled to a processor so that the processor can read and write information from / to the storage medium. In an alternative, a storage medium may be integrated into a processor. The processor and the storage medium may reside in an ASIC. The ASIC may reside in a user terminal. In an alternative, the processor and the storage medium may reside in a user terminal as discrete components.

[0054] In one or more exemplary embodiments, the functions described may be implemented in hardware, software, firmware, or any combination thereof. If implemented as a computer program product in software, each function may be stored on or transmitted by a computer-readable medium as one or more instructions or codes. Computer-readable media include both computer storage media and communication media, including any media that facilitates the transfer of a computer program from one place to another. Storage media may be any available media that can be accessed by a computer. As an example and not limitation, such a computer-readable medium may include RAM, ROM, EEPROM, CD-ROM or other optical disk storage, disk storage or other magnetic storage device, or any other medium that can be used to carry or store the desired program code in the form of an instruction or data structure and can be accessed by a computer. Any connection is also properly referred to as a computer-readable medium. For example, if the software is transmitted from a Web site, server, or other remote source using a coaxial cable, fiber optic cable, twisted pair, digital subscriber line (DSL), or wireless technologies such as infrared, radio, and microwaves, the coaxial cable, fiber optic cable, twisted pair, DSL, or wireless technologies such as infrared, radio, and microwaves are included in the definition of the medium. Disk and disc as used herein include compact disc (CD), laser disc, optical disc, digital versatile disc (DVD), floppy disk and Blu-ray disc, where disks usually reproduce data magnetically, while discs reproduce data optically with lasers. Combinations of the above should also be included within the scope of computer-readable media.

[0055] The embodiments described above provide a detailed description of the technical solutions and beneficial effects of the present invention. It should be understood that the above are only specific embodiments of the present invention and are not intended to limit the present invention. Any modifications, supplements and equivalent substitutions made within the scope of the principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for predicting the affinity of an antibody mutant to an antigen, characterized in that: The following steps are involved: (1) Collect affinity experimental data of antibodies and antibody-antigen interactions and construct a fine-tuning dataset; (2) Use the fine-tuning dataset to train the UniRep model to obtain the affinity prediction model between antibodies and antigens; (3) Mutate key sites in the antibody sequence to obtain antibody mutants; use the affinity prediction model to predict the affinity between the antibody mutants and the antigen.

2. The method for predicting the affinity between an antibody mutant and an antigen according to claim 1, characterized in that: The fine-tuning data set includes the amino acid sequence of the antibody and the affinity measurement value of the interaction between the antibody and the antigen.

3. The method for predicting affinity between antibody mutants and antigens according to claim 1, characterized in that: Step (2) includes: (2-1) Divide the fine-tuning dataset into training set and test set; (2-2) The training set is used to supervise the learning of the UniRep model, and the UniRep model parameters are gradually optimized by minimizing the error between the affinity prediction value and the actual affinity measurement value; after the training is completed, the affinity prediction model is obtained; (2-3) The test set is used to evaluate the model’s predictive ability on unseen data.

4. The method for predicting affinity between antibody mutants and antigens according to claim 3, characterized in that: In step (2-1), the fine-tuning dataset is divided into K subsets of equal size using the K-fold cross-validation strategy. One subset is used as the test set each time, and the remaining K-1 subsets are used as training sets.

5. The method for predicting affinity between antibody mutants and antigens according to claim 3, characterized in that: Step (2-2) includes: (2-2i) Antibody X i The amino acid sequence of antibody X was input into the UniRep model, and the UniRep model extracted i The sequence features of antibody X are predicted based on the sequence features. i Affinity of interaction with antigen; formula: ； in, is the ith antibody X i UniRep(X) represents the predicted affinity value of antibody X by the UniRep model. i The amino acid sequence is extracted from the sequence features; f is the regression function, which represents the mapping relationship from sequence features to affinity; ϵ is the prediction error term; (2-2ii) The mean square error is used as the loss function to calculate the error between the predicted affinity value and the measured affinity value: ； Among them, Y i is the actual affinity value measured in the experiment, is the predicted value of the model, and N is the number of samples in the test set; (2-2iii) During the model training process, the gradient of the loss function to the model parameters is calculated through the back propagation algorithm, and the model parameters are updated using the gradient descent algorithm; the specific optimization formula is as follows: ； in, θ t is the current model parameter; η is the learning rate; is the gradient of the loss function with respect to the model parameters; (2-2iv) Through continuous iteration and optimization, an affinity prediction model is obtained.

6. The method for predicting affinity between antibody mutants and antigens according to claim 5, characterized in that: In step (2-2i), the UniRep model extracts antibody X i The sequence features include: i The amino acid sequence is converted into an amino acid feature vector; The amino acid features of each time step are processed step by step through the LSTM network to generate the context-dependent features of the amino acid sequence; the formula is expressed as: ； Among them, h t represents the hidden state of the LSTM network at time step t; x t represents the input features of the antibody amino acid sequence at time step t; h t-1 represents the hidden state of the previous time step t-1; The hidden state of the last time step is the context-dependent feature of the amino acid sequence, which is antibody X i sequence characteristics.

7. The method for predicting affinity between antibody mutants and antigens according to claim 1, characterized in that: In step (3), the mutation at the key site of the antibody sequence includes: obtaining antibody mutants by single point mutation or multiple point mutation in the complementarity determining region of the antibody sequence.

8. The method for predicting affinity between antibody mutants and antigens according to claim 1, characterized in that: In step (3), the affinity prediction model is used to predict the affinity between the antibody mutant and the antigen, including: inputting the amino acid sequence of the antibody mutant into the affinity prediction model, the affinity prediction model extracts the sequence features of the antibody mutant, and predicts the affinity of the interaction between the antibody mutant and the antigen based on the sequence features, the formula is: ； in, Is the affinity prediction model for antibody mutant X mut The affinity prediction value of UniRep(X mut ) represents the sequence features of the antibody mutants extracted by the affinity prediction model; f is a regression function that directly maps the extracted features to affinity values.

9. The method for predicting affinity between antibody mutants and antigens according to claim 1, characterized in that: The antibody is an anti-HER2 antibody; the antigen is a HER2 antigen.

10. A system for predicting the affinity of antibody mutants to antigens, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: The memory also stores an affinity prediction model obtained by the prediction method according to any one of claims 1 to 9, and the processor implements the following steps when executing the computer program: The amino acid sequence of the antibody mutant is input into the affinity prediction model, and the predicted value of the affinity between the antibody mutant and the antigen is output.

Images