Antigen-antibody affinity prediction method, system and equipment

Abstract

The invention relates to the technical field of artificial intelligence, in particular to an antigen-antibody affinity prediction method, system and equipment, and the method comprises the following steps: carrying out feature extraction on amino acid sequence data of an antigen and an antibody, and generating a preliminary compound three-dimensional conformation through a prediction model; fusing and screening the plurality of preliminary compound three-dimensional conformations, and determining an intermediate compound three-dimensional conformation; on the basis of molecular dynamics simulation and reinforcement learning algorithms, optimizing a binding interface of the three-dimensional conformation of the intermediate compound to obtain a final optimized conformation; performing energy decomposition on the basis of a molecular force field principle to obtain key energy parameters, and predicting a deep neural network through affinity on the basis of the key energy parameters and energy characteristics determined by molecular dynamics simulation to obtain combined free energy; and when the combined free energy is higher than the judgment threshold value, optimization is carried out again, and the combined free energy is determined. Therefore, on the basis of realizing full-process automation, the accuracy and reliability of prediction can be obviously improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an antigen-antibody affinity prediction method, system and device. BACKGROUND

[0002] The binding affinity between antigens and antibodies is a core indicator for evaluating the efficacy of antibody drugs, the protection effect of vaccines, and understanding the mechanism of immune response, which can generally be characterized by binding free energy. Traditional experimental methods, such as surface plasmon resonance technology or isothermal titration calorimetry, can accurately measure the affinity, but generally have problems such as high cost, long cycle, large sample consumption, and are difficult to meet the needs of high-throughput screening and rapid iteration. SUMMARY

[0003] Therefore, the purpose of the present application is to provide an antigen-antibody affinity prediction method, system and device to solve the problems of high cost, long cycle and large sample consumption in the prior art.

[0004] To achieve the above purpose, the present application adopts the following technical solutions: In a first aspect, the present application provides an antigen-antibody affinity prediction method, comprising: extracting features from amino acid sequence data of antigens and antibodies, and generating preliminary complex three-dimensional conformations based on the feature extraction results and a preset prediction model; fusing and screening a plurality of preliminary complex three-dimensional conformations to determine intermediate complex three-dimensional conformations; optimizing the binding interface of the intermediate complex three-dimensional conformations based on molecular dynamics simulation and reinforcement learning algorithm to obtain a final optimized conformation; based on the principle of molecular force field, energy decomposition is performed on the final optimized conformation to obtain key energy parameters, and based on the key energy parameters and energy characteristics determined by molecular dynamics simulation, the binding free energy is obtained through a preset affinity prediction deep neural network; comparing the binding free energy with a preset judgment threshold, if the binding free energy is higher than the judgment threshold, re-optimizing the binding interface of the intermediate complex three-dimensional conformations and determining the binding free energy, otherwise outputting the binding free energy.

[0005] Further, in some embodiments of the present application, the feature extraction from the amino acid sequence data of antigens and antibodies, and the generation of preliminary complex three-dimensional conformations based on the feature extraction results and a preset prediction model, comprises: performing multiple sequence alignment processing on the amino acid sequence data to extract sequence conservation features; converting the amino acid sequence data into a high-dimensional feature vector based on a pre-trained language model; Based on the sequence conservation features and the high-dimensional feature vector, monomer structure prediction is performed on the antigen and the antibody respectively by a preset monomer prediction model, to generate a plurality of candidate structures; The candidate structures are input into a preset complex prediction model to generate the preliminary complex three-dimensional conformation.

[0006] Further, in some embodiments of the present application, the fusion and screening of the plurality of preliminary complex three-dimensional conformations to determine the intermediate complex three-dimensional conformation, comprises: Comparing the structures of the plurality of preliminary complex three-dimensional conformations, and merging the preliminary complex three-dimensional conformations based on the comparison results to obtain a fusion conformation set; wherein the plurality of preliminary complex three-dimensional conformations are output by a plurality of complex prediction models or output multiple times by one complex prediction model; Clustering and screening the complex three-dimensional conformations in the fusion conformation set to obtain the intermediate complex three-dimensional conformation.

[0007] Further, in some embodiments of the present application, the binding interface of the intermediate complex three-dimensional conformation is optimized based on molecular dynamics simulation and reinforcement learning algorithm to obtain the final optimized conformation, comprising: Performing coarse-grained molecular dynamics simulation on the intermediate complex three-dimensional conformation to obtain coarse-grained simulation results; Extracting the binding interface data of the coarse-grained simulation results; Reconstructing the side chain of the binding interface residues, and adjusting the rotation angle and conformation of the residues based on the policy gradient reinforcement learning algorithm to obtain the optimization result; Performing all-atom molecular dynamics refinement on the optimization result to obtain the final optimized conformation.

[0008] Further, in some embodiments of the present application, based on the principle of molecular force field, the energy decomposition is performed on the final optimized conformation to obtain key energy parameters, and based on the key energy parameters and the energy features determined by the molecular dynamics simulation, the binding free energy is obtained by a preset affinity prediction deep neural network, comprising: Obtaining the energy features obtained based on coarse-grained molecular dynamics simulation and all-atom molecular dynamics refinement; Inputting the key energy parameters and the energy features into the affinity prediction deep neural network to predict the binding free energy.

[0009] Further, in some embodiments of the present application, the key energy parameters include van der Waals energy, electrostatic energy and solvation free energy.

[0010] Further, in some embodiments of the present application, further comprising: performing uncertainty evaluation on the predicted binding free energy to determine a corresponding confidence interval.

[0011] Further, in some embodiments of the present application, further comprising: generating single-point and multi-point mutation conformations of the binding interface residues; calculating the difference in binding free energy before and after mutation for each mutation conformation, and analyzing the influence of mutation on affinity based on the difference in binding free energy.

[0012] In a second aspect, the present application provides an antigen-antibody affinity prediction system, comprising: a data input and structure prediction module for feature extraction on amino acid sequence data of the antigen and the antibody, and generating a preliminary complex three-dimensional conformation based on the feature extraction result and a preset prediction model; a conformation fusion and screening module for fusing and screening a plurality of preliminary complex three-dimensional conformations to determine an intermediate complex three-dimensional conformation; a binding interface optimization module for optimizing the binding interface of the intermediate complex three-dimensional conformation based on molecular dynamics simulation and reinforcement learning algorithm to obtain a final optimized conformation; a cross-scale affinity calculation module for energy decomposition based on the principle of molecular force field for the final optimized conformation to obtain key energy parameters, and based on the key energy parameters and energy characteristics determined by molecular dynamics simulation, obtaining the binding free energy through a preset affinity prediction deep neural network; and comparing the binding free energy with a preset judgment threshold, if the binding free energy is higher than the judgment threshold, it is judged that the binding interface of the intermediate complex three-dimensional conformation needs to be optimized and the binding free energy needs to be determined, otherwise the binding free energy is output.

[0013] In a third aspect, the present application provides an antigen-antibody affinity prediction device, characterized in that it comprises a processor and a memory, wherein the processor is connected with the memory: wherein the processor is used to call and execute the program stored in the memory; the memory is used to store the program, and the program is used to execute at least the antigen-antibody affinity prediction method.

[0014] The application relates to the technical field of artificial intelligence, in particular to an antigen-antibody affinity prediction method, system and device. BRIEF DESCRIPTION OF DRAWINGS

[0015] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of these drawings.

[0016] Figure 1 is a flowchart of the antigen-antibody affinity prediction method provided by the embodiment of the present application.

[0017] Figure 2 is a structural schematic diagram of the antigen-antibody affinity prediction system provided by the embodiment of the present application.

[0018] Figure 3 is a structural schematic diagram of the antigen-antibody affinity prediction device provided by the embodiment of the present application. DETAILED DESCRIPTION

[0019] In order to make the objects, technical solutions and advantages of the present application clearer, the technical solutions of the present application will be described in detail below. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort fall within the scope of the present application.

[0020] Figure 1 is a flowchart of the antigen-antibody affinity prediction method provided by the embodiment of the present application, please refer to Figure 1 , the present embodiment can include the following steps: S101, feature extraction is performed on the amino acid sequence data of the antigen and the antibody, and a preliminary complex three-dimensional conformation is generated based on the feature extraction result and a preset prediction model.

[0021] Specifically, the amino acid sequence data of the antigen and the antibody can be derived from gene sequencing, public databases or artificial design, etc. In the present application, after feature extraction is performed on the amino acid sequence data of the antigen and the antibody, the feature extraction result is taken as the input of the preset prediction model, so as to generate and output the preliminary complex three-dimensional conformation through the prediction model.

[0022] S102, a plurality of preliminary complex three-dimensional conformations are fused and screened to determine an intermediate complex three-dimensional conformation.

[0023] Specifically, in the present application, a plurality of preliminary complex three-dimensional conformations can be output by a plurality of complex prediction models or output multiple times by one complex prediction model, and then the plurality of preliminary complex three-dimensional conformations are fused and screened to determine the intermediate complex three-dimensional conformation.

[0024] S103, based on molecular dynamics simulation and reinforcement learning algorithm, the binding interface of the intermediate complex three-dimensional conformation is optimized to obtain a final optimized conformation.

[0025] Specifically, in the present application, the binding interface data is determined by performing molecular dynamics simulation on the intermediate complex three-dimensional conformation, and on this basis, the final optimization result is obtained by optimizing the binding interface, which is used for subsequent calculation of the binding free energy.

[0026] S104, based on the principle of molecular force field, the energy of the final optimized conformation is decomposed to obtain key energy parameters, and based on the key energy parameters and the energy characteristics determined by the molecular dynamics simulation, the binding free energy is obtained through a preset affinity prediction deep neural network.

[0027] S105, the binding free energy is compared with a preset judgment threshold, if the binding free energy is higher than the judgment threshold, the binding interface of the intermediate complex three-dimensional conformation is re-optimized and the binding free energy is determined, otherwise the binding free energy is output.

[0028] The antigen-antibody affinity prediction method provided in the application can significantly improve the accuracy and reliability of prediction on the basis of realizing full-process automation.

[0029] Further, in some embodiments of the application, the amino acid sequence data of the antigen and the antibody is subjected to feature extraction, and based on the feature extraction result and a preset prediction model, a preliminary complex three-dimensional conformation is generated, which can specifically include: first, the amino acid sequence data is subjected to multiple sequence alignment processing to extract sequence conservation features; then the amino acid sequence data is converted into a high-dimensional feature vector based on a pre-trained language model; then based on the sequence conservation features and the high-dimensional feature vector, the monomer structure of the antigen and the antibody is predicted respectively by a preset monomer prediction model to generate a plurality of candidate structures; and the candidate structures are input into a preset complex prediction model to generate a preliminary complex three-dimensional conformation.

[0030] In actual application, the monomer prediction model can be constructed by using a deep generative modeling network to predict the monomer structure of the antigen and the antibody respectively to generate a plurality of candidate structures to obtain a candidate structure set. Then the predicted structures of the antigen and the antibody in the candidate structure set are input into the complex prediction model to generate a plurality of preliminary complex three-dimensional conformations (which can be output by a plurality of complex prediction models or output by one complex prediction model multiple times) to obtain a preliminary complex three-dimensional conformation set. It should be noted that the confidence score corresponding to the preliminary complex three-dimensional conformation can also be obtained by the complex prediction model.

[0031] On this basis, the multiple preliminary complex three-dimensional conformations are subjected to structure comparison, and based on the comparison result, the preliminary complex three-dimensional conformations are merged to obtain a fusion conformation set; and the complex three-dimensional conformations in the fusion conformation set are subjected to clustering screening to obtain intermediate complex three-dimensional conformations.

[0032] Specifically, first, the multiple preliminary complex three-dimensional conformations are subjected to structure comparison, and similar conformations are merged to obtain a fusion conformation set. Then, based on the confidence score obtained by the above prediction model or the conformation clustering center degree information, the intermediate complex three-dimensional conformations are screened out as the input for subsequent optimization.

[0033] Further, in some embodiments of the present application, the binding interface of the intermediate complex three-dimensional conformation is optimized based on molecular dynamics simulation and reinforcement learning algorithm, and the final optimized conformation is obtained, which can specifically include: First, coarse-grained molecular dynamics simulation is performed on the intermediate complex three-dimensional conformation to obtain coarse-grained simulation results, so as to realize rapid relaxation of the global conformation.

[0034] Then, the binding interface data of the coarse-grained simulation results is extracted.

[0035] Then, side chain reconstruction is performed on the binding interface residues, and the rotation angle and conformation of the residues are adjusted based on the policy gradient reinforcement learning algorithm to obtain the optimization result. In actual application, the optimization can be performed with the goal of maximizing the interface electrostatic and hydrophobic complementarity score.

[0036] Finally, the optimization result is refined by all-atom molecular dynamics to obtain the final optimized conformation.

[0037] Further, in some embodiments of the present application, based on the principle of molecular force field, the energy of the final optimized conformation is decomposed to obtain key energy parameters, and based on the key energy parameters and the energy characteristics determined by the molecular dynamics simulation, the binding free energy is obtained through the preset affinity prediction deep neural network, which can specifically include: obtaining the energy characteristics based on the coarse-grained molecular dynamics simulation and the all-atom molecular dynamics refinement; and inputting the key energy parameters and the energy characteristics into the affinity prediction deep neural network to predict the binding free energy.

[0038] Specifically, for the final optimized conformation, molecular force field can be used for energy decomposition to obtain key energy parameters such as van der Waals energy, electrostatic energy and solvation free energy; then the energy characteristics determined by the above coarse-grained molecular dynamics simulation and all-atom molecular dynamics refinement are input into the preset affinity prediction deep neural network, and the network outputs the binding free energy. In addition, the uncertainty of the binding free energy can be evaluated by Monte Carlo Dropout, Bootstrap sampling and other methods, and the confidence interval is output.

[0039] On this basis, after obtaining the binding free energy, it is compared with the preset judgment threshold value, if the binding free energy is higher than the judgment threshold value, it indicates that the affinity is insufficient, then the binding interface of the intermediate complex three-dimensional conformation is optimized and the binding free energy is determined again (the process and principle steps are the same as those in the above embodiments, and will not be described here), otherwise the binding free energy is output, so as to ensure the reliability of the final binding free energy.

[0040] In addition, in some embodiments of the present application, the antigen-antibody affinity prediction method provided by the present application further comprises: generating single-point and multi-point mutation conformations of the binding interface residues; calculating the difference in binding free energy before and after mutation for each mutation conformation, and analyzing the influence of mutation on affinity based on the difference in binding free energy.

[0041] For example, the single-point and multi-point mutation conformations of the binding interface residues (such as amino acid scanning, immune escape mutation set) can be automatically generated by a preset system; then, the above-mentioned binding interface is used for optimization and calculation of free binding energy in each mutation conformation, so as to obtain the binding free energy before and after mutation, and further calculate the difference in binding free energy before and after mutation. Then, the difference is used to analyze the influence of mutation on affinity, such as generating a mutation sensitivity heat map for analyzing the sensitivity of binding affinity to different residue mutations.

[0042] Based on the same inventive concept, the present application also provides an antigen-antibody affinity prediction system, Figure 2 which is a structural schematic diagram of the antigen-antibody affinity prediction system provided by the present application, as shown in the figure, the system comprises: Figure 2 a data input and structure prediction module 11 for feature extraction on amino acid sequence data of the antigen and the antibody, and generating a preliminary complex three-dimensional conformation based on the feature extraction result and a preset prediction model.

[0043] In actual application, the module can include a sequence input and feature extraction sub-module and a three-dimensional structure prediction sub-module; wherein the amino acid sequence data of the antigen and the antibody can be input through the sequence input and feature extraction sub-module, and then the module is used for feature extraction and coding by using multiple sequence alignment and a pre-trained language model; and the three-dimensional structure of the complex is predicted through the three-dimensional structure prediction sub-module, and the prediction confidence is output.

[0044] a conformation fusion and screening module 12 for fusing and screening a plurality of preliminary complex three-dimensional conformations to determine an intermediate complex three-dimensional conformation. In actual application, the module can screen out the intermediate complex three-dimensional conformation by using RMSD clustering or an energy function for a plurality of preliminary complex three-dimensional conformations.

[0045] a binding interface optimization module 13 for optimizing the binding interface of the intermediate complex three-dimensional conformation based on molecular dynamics simulation and reinforcement learning algorithm to obtain a final optimized conformation, including the coarse-grained molecular dynamics simulation, side chain reconstruction of the binding interface residues, optimization of the residue rotation angle and conformation guided by reinforcement learning, and all-atom molecular dynamics refinement mentioned in the above-mentioned method embodiments.

[0046] ​The cross-scale affinity calculation module 14 is configured to perform energy decomposition on the final optimized conformation based on the principle of molecular force field, obtain key energy parameters, and obtain the binding free energy by a preset affinity prediction deep neural network based on the key energy parameters and energy characteristics determined by the molecular dynamics simulation; and compare the binding free energy with a preset judgment threshold value, if the binding free energy is higher than the judgment threshold value, it is determined that the binding interface of the intermediate complex three-dimensional conformation needs to be optimized and the binding free energy is determined, otherwise the binding free energy is output. Of course, this module can also be used only to obtain the function of obtaining the binding free energy, and other separate modules are used to realize the function of comparing the binding free energy with the judgment threshold value, and the connection with other modules is used to realize the feedback closed loop.

[0047] In addition, in some embodiments of the present application, the system further comprises a mutation sensitivity analysis module configured to trigger automatic generation of mutant conformations based on the antibody or antigen mutant sequence, and calculate the change value of the binding free energy before and after the mutation, to generate a mutation site sensitivity heat map.

[0048] In actual application, the above system can also set specific input content according to demand, for example, in addition to outputting the affinity prediction result, i.e. the binding free energy, it can also output related three-dimensional structure files, confidence interval information and mutation sensitivity heat map, etc. for relevant personnel to check and analyze.

[0049] In addition, it should be noted that the above-mentioned three-dimensional structure prediction sub-module and mutation sensitivity analysis module can be run in parallel on different computing nodes to improve large-scale prediction efficiency, and the system can be deployed on a high-performance computing cluster or a GPU-accelerated workstation, etc.

[0050] As to the device in the above-mentioned embodiments, the specific manner in which each module performs an operation has been described in detail in the embodiments related to the method, and will not be described in detail here.

[0051] The antigen-antibody affinity prediction system provided by the present application can realize complete automation from sequence data input to affinity prediction, without manual docking and energy calculation configuration, greatly reducing labor cost and operation time; at the same time, by fusing the prediction results of multiple prediction models and screening the uncertainty clusters, the stability and robustness of structure prediction are improved; and by using reinforcement learning, the binding interface conformation can be actively optimized to improve the accuracy of prediction; and by using the data of coarse-grained and all-atom simulation, the binding free energy is predicted by deep neural network, to balance the accuracy and calculation speed; and by mutation sensitivity analysis, the influence of mutation on affinity can be quickly evaluated, thereby providing support for immune escape risk assessment and antibody resistance research, etc.

[0052] Based on the same inventive concept, the application further provides an antigen-antibody affinity prediction device for implementing the method embodiments. Figure 3 is a structural schematic diagram of the antigen-antibody affinity prediction device provided by the embodiment of the application, as shown in the figure, the antigen-antibody affinity prediction device of the embodiment comprises a processor 21 and a memory 22, the processor 21 is connected with the memory 22. Among them, the processor 21 is used to call and execute the program stored in the memory 22; the memory 22 is used to store the program, which is at least used to execute the antigen-antibody affinity prediction method in the above embodiment. Figure 3

[0053] The specific implementation of the antigen-antibody affinity prediction device provided by the embodiment of the application can refer to the implementation of the antigen-antibody affinity prediction method of any of the above embodiments, which will not be repeated here.

[0054] It can be understood that the same or similar parts in the above embodiments can be mutually referred to, and the contents not described in detail in some embodiments can be referred to the same or similar contents in other embodiments.

[0055] It should be noted that in the description of the application, the terms "first", "second", etc. are only for the purpose of description, and cannot be understood as indicating or implying relative importance. In addition, in the description of the application, unless otherwise specified, the meaning of "a plurality of" is at least two.

[0056] Any process or method descriptions in flow charts or otherwise described herein represents an example of embodiments of the present application that can be embodied in code means that include one or more steps for accomplishing a specified logical function or process. The scope of embodiments of the present application encompassed by the preferred embodiments also include hardware and software configured to perform the specified functions of examples of embodiments of the present application in no particular order, including essentially simultaneous functions, as well as in reverse order. It should be understood that the scope of the embodiments of the present application encompasses all possible combinations that can be derived from the described examples of embodiments of the present application.

[0057] It should be understood that the parts of the present application can be realized by hardware, software, firmware or their combination. In the above implementation, a plurality of steps or methods can be realized by software or firmware stored in the memory and executed by a suitable instruction execution system. For example, if realized by hardware, and as in another embodiment, it can be realized by any one or combination of the following technologies known in the art: discrete logic circuit with logic gate circuit for implementing logic function on data signal, special integrated circuit with suitable combination logic gate circuit, programmable gate array (PGA), field programmable gate array (FPGA) and the like.

[0058] ​Those skilled in the art can understand that all or part of the steps of the method carried out by the above-mentioned embodiments can be instructed by a program to the relevant hardware, and the program can be stored in a computer readable storage medium. When the program is executed, it includes one of the steps of the method embodiment or a combination thereof.

[0059] In addition, each functional unit in each embodiment of the present application can be integrated into one processing module, or each unit can exist physically independently, or two or more units can be integrated into one module. The integrated module can be realized in the form of hardware or in the form of a software functional module. When the integrated module is realized in the form of a software functional module and sold or used as an independent product, it can also be stored in a computer readable storage medium.

[0060] The storage medium mentioned above can be a read-only memory, a magnetic disk or an optical disk, etc.

[0061] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example" or "some examples" means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative description of the above terms does not necessarily mean the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described can be combined in any one or more embodiments or examples in a suitable manner.

[0062] Although the embodiments of the present application have been shown and described above, it can be understood that the above-mentioned embodiments are exemplary and cannot be understood as limiting the present application. Those skilled in the art can make changes, modifications, replacements and variations to the above-mentioned embodiments within the scope of the present application.

Claims

1. A method for predicting antigen-antibody affinity, characterized in that, include: Features are extracted from the amino acid sequence data of antigens and antibodies, and a preliminary three-dimensional conformation of the complex is generated based on the feature extraction results and a pre-set prediction model. Multiple preliminary three-dimensional conformations of the complex were fused and screened to determine the three-dimensional conformation of the intermediate complex. Based on molecular dynamics simulations and reinforcement learning algorithms, the binding interface of the three-dimensional conformation of the intermediate complex is optimized to obtain the final optimized conformation. Based on the principle of molecular force field, energy decomposition is performed on the final optimized conformation to obtain key energy parameters. Based on the key energy parameters and the energy characteristics determined by molecular dynamics simulation, the binding free energy is obtained through a preset affinity prediction deep neural network. The binding free energy is compared with a preset judgment threshold. If the binding free energy is higher than the judgment threshold, the binding interface of the three-dimensional conformation of the intermediate complex is re-optimized and the binding free energy is determined. Otherwise, the binding free energy is output.

2. The antigen-antibody affinity prediction method according to claim 1, characterized in that, The process involves feature extraction from the amino acid sequence data of the antigen and antibody, and based on the feature extraction results and a pre-defined prediction model, generating a preliminary three-dimensional conformation of the complex, including: Multiple sequence alignment was performed on amino acid sequence data to extract sequence conservation features; Pre-trained language models are used to convert amino acid sequence data into high-dimensional feature vectors. Based on the sequence conservation features and high-dimensional feature vectors, the monomer structure of antigens and antibodies is predicted by a preset monomer prediction model, generating multiple candidate structures. The candidate structure is input into a preset complex prediction model to generate the preliminary three-dimensional conformation of the complex.

3. The antigen-antibody affinity prediction method according to claim 2, characterized in that, The process of fusing and screening multiple preliminary three-dimensional conformations of the complex to determine the three-dimensional conformation of the intermediate complex includes: Multiple preliminary three-dimensional conformations of the complex are structurally compared, and the preliminary three-dimensional conformations of the complex are merged based on the comparison results to obtain a fused conformation set; wherein, the multiple preliminary three-dimensional conformations of the complex are output by multiple complex prediction models or by a single complex prediction model multiple times; Clustering and screening of the three-dimensional conformations of the complex in the fusion conformation set yields the three-dimensional conformation of the intermediate complex.

4. The antigen-antibody affinity prediction method according to claim 3, characterized in that, The binding interface of the intermediate complex's three-dimensional conformation is optimized based on molecular dynamics simulations and reinforcement learning algorithms to obtain the final optimized conformation, including: Coarse-grained molecular dynamics simulations were performed on the three-dimensional conformation of the intermediate complex to obtain coarse-grained simulation results; Extract the interface data combining the coarse-grained simulation results; The optimization results were obtained by combining interface residues for side chain reconstruction and using a policy gradient-based reinforcement learning algorithm to adjust the rotation angle and conformation of residues. The optimization results were refined using all-atom molecular dynamics to obtain the final optimized conformation.

5. The antigen-antibody affinity prediction method according to claim 4, characterized in that, The process involves energy decomposition of the final optimized conformation based on molecular force field principles to obtain key energy parameters. Based on these key energy parameters and energy characteristics determined by molecular dynamics simulations, a pre-defined affinity prediction deep neural network is used to obtain the binding free energy, including: Obtain energy characteristics based on coarse-grained molecular dynamics simulations and all-atom molecular dynamics refinement; The key energy parameters and energy features are input into the affinity prediction deep neural network to predict the binding free energy.

6. The antigen-antibody affinity prediction method according to claim 1, characterized in that, The key energy parameters include van der Waals energy, electrostatic energy, and solvation free energy.

7. The antigen-antibody affinity prediction method according to claim 5, characterized in that, Also includes: An uncertainty assessment is performed on the predicted binding free energy to determine the corresponding confidence interval.

8. The antigen-antibody affinity prediction method according to claim 5, characterized in that, Also includes: Generate single-point and multi-point mutant conformations of binding interface residues; For each mutated conformation, the difference in binding free energy before and after the mutation is calculated, and the effect of the mutation on affinity is analyzed based on the difference in binding free energy.

9. An antigen-antibody affinity prediction system, characterized in that, include: The data input and structure prediction module is used to extract features from the amino acid sequence data of antigens and antibodies, and generate a preliminary three-dimensional conformation of the complex based on the feature extraction results and the preset prediction model. The conformation fusion and screening module is used to fuse and screen multiple preliminary three-dimensional conformations of the complex to determine the three-dimensional conformation of the intermediate complex. The interface optimization module is used to optimize the binding interface of the three-dimensional conformation of the intermediate complex based on molecular dynamics simulation and reinforcement learning algorithm to obtain the final optimized conformation. The cross-scale affinity calculation module is used to perform energy decomposition on the final optimized conformation based on the principle of molecular force field to obtain key energy parameters, and based on the key energy parameters and energy characteristics determined by molecular dynamics simulation, to obtain the binding free energy through a preset affinity prediction deep neural network. The binding free energy is compared with a preset judgment threshold. If the binding free energy is higher than the judgment threshold, it is determined that the binding interface of the three-dimensional conformation of the intermediate complex needs to be optimized and the binding free energy needs to be determined. Otherwise, the binding free energy is output.

10. An antigen-antibody affinity prediction device, characterized in that, It includes a processor and a memory, wherein the processor is connected to the memory: The processor is used to call and execute the program stored in the memory; The memory is used to store the program, which is at least used to execute the antigen-antibody affinity prediction method according to any one of claims 1-8.

Images