Antibody screening method and system based on differential evolution algorithm and artificial intelligence model

Abstract

The invention provides an antibody screening method and system based on a differential evolution algorithm and an artificial intelligence model. The method comprises the following steps: S1, acquiring data of an antibody-antigen compound sequence and affinity; s2, encoding the antibody and antigen sequences by using a language model, and converting the encoded antibody and antigen sequences into a matrix M; s3, training a machine learning model by using the matrix M and the affinity parameters to obtain a trained optimal machine learning model; s4, predicting the antigen affinity of the antibody by utilizing the trained optimal machine learning model; and S5, carrying out affinity optimization on the antibody sequence based on a differential evolution algorithm, and carrying out further filtration by predicting the druggability of the antibody so as to screen the antibody with high affinity and good drug properties.

Description

Technical Field

[0001] The present invention relates to the technical field of bioinformatics, and in particular to an antibody screening method and system based on a differential evolution algorithm and an artificial intelligence model. Background Art

[0002] In the field of biomedicine, antibodies, as an important biomolecular tool, are widely used in disease diagnosis, treatment and research. The affinity of an antibody is a key indicator of its ability to bind to the target molecule. Antibodies with high affinity usually have better efficacy and wider application prospects.

[0003] Affinity refers to the specific binding strength between the antibody and the corresponding epitope in the antigen, which can be expressed by the binding constant Kd. The smaller the Kd, the tighter the binding. The size of affinity is affected by many factors, including the structure, charge, hydrophobicity and other factors of the antibody and antigen, as well as environmental factors such as temperature and pH.

[0004] Affinity maturation refers to the process in which the antigen binds to the primary antibody and the recognized B cells enter the germinal center of the lymph node. Then, the antibody undergoes somatic mutation, selective retention, competitive selection and other processes, so that antibodies with higher affinity are selected and produced.

[0005] Traditional methods for predicting antibody affinity are usually based on physical, empirical and knowledge-based scoring functions. Physical scoring functions consider the physicochemical properties of antibody-antigen binding, such as binding energy, electrostatic interactions, etc. Empirical scoring functions rely on known antibody-antigen structures and experimental data to evaluate affinity by summarizing empirical rules and knowledge. Knowledge-based scoring functions usually use statistical methods to determine the frequency of occurrence of interacting residue pairs or atom pairs in protein complexes, and then calculate the interaction energy through Boltzmann distribution.

[0006] In vitro affinity maturation is a molecular evolution-based method used to enhance the binding ability and specificity of antibodies. By introducing genetic mutations in vitro, antibodies with higher affinity and specificity are screened. Commonly used methods include targeted mutagenesis and random mutagenesis.

[0007] Computational antibody affinity maturation usually requires understanding the amino acid residues that play a role in the antibody-antigen binding interaction based on the existing antibody structure through methods such as molecular docking, and then using molecular dynamics simulation, scoring function and other methods to evaluate the effects of different amino acid substitutions on the antibody-antigen binding free energy. Finally, the optimal amino acid substitution scheme is selected and experimentally verified.

[0008] Traditional methods for predicting antibody affinity often have shortcomings such as limited accuracy, high computational complexity, and long time consumption, which limit the accuracy, comprehensiveness, and applicability of traditional methods.

[0009] The main disadvantage of in vitro affinity maturation is that the size of the variant library is limited, and it is difficult to generate all possible combinations of CDR mutations. In addition, even if CDR residues are replaced, there is no guarantee that the binding affinity to the antigen will be improved, so it is difficult to determine which residues should be mutagenized.

[0010] The effectiveness of traditional computational affinity maturation methods is limited by the accuracy of simulation and scoring functions. The difficulty of CDR-H3 modeling, the accuracy of rigid body docking and the precision of molecular dynamics simulation force fields limit the effectiveness of computationally assisted affinity maturation.

[0011] Current traditional affinity maturation methods are often limited by the complexity and diversity of the design space, and are unable to conduct comprehensive searches of large-scale antibody sequences, nor can they fully utilize the knowledge and experience of domain experts. In addition, the multiple objectives and constraints in antibody design, such as affinity, specificity, stability, etc., need to be considered comprehensively, making the optimization problem more complicated.

[0012] Patent document CN102132143B (application number: 200980133062.7) discloses a method for screening high-affinity antibodies, comprising the steps of: a) providing a plurality of hybridoma cells, each cell expressing an antibody, b) determining the time-dependent amount of binding of the antibody to each antigen by surface plasmon resonance at different temperatures and different antibody concentrations, c) based on equations (II) to (XIII), using the time-dependent amount determined in b) to calculate at least the thermodynamic parameters (i) standard association binding entropy formula (A), (ii) standard dissociation binding entropy formula (B), (iii) standard binding entropy (ΔS°), (iv) free standard binding enthalpy (ΔG°), (v) standard dissociation free binding enthalpy formula (C), (vi) standard association free binding enthalpy formula (D), (vii) -TΔS°, (viii) dissociation rate constant kd, (ix) equilibrium binding constant KD, and (x) association rate constant Ka, d) selecting hybridoma cells that produce such antibodies, wherein the antibodies have at least two of the following: i) a standard association binding entropy of less than 10 J / K*mol, ii) an absolute standard dissociation binding entropy of 100 J / mol*K or more, iii) an absolute standard binding entropy of 100 J / mol*K or more, e) producing the antibody by culturing the selected cells under conditions suitable for expression of the antibody, and recovering the antibody from the cells and / or the culture medium. Summary of the invention

[0013] In view of the defects in the prior art, the object of the present invention is to provide an antibody screening method and system based on differential evolution algorithm and artificial intelligence model.

[0014] According to the present invention, an antibody screening method based on a differential evolution algorithm and an artificial intelligence model comprises:

[0015] Step S1: obtaining antibody-antigen complex sequence and affinity data;

[0016] Step S2: Encode the antibody and antigen sequences using a language model and convert them into a matrix M;

[0017] Step S3: using the matrix M and affinity parameters to train the machine learning model to obtain the optimal machine learning model after training;

[0018] Step S4: using the trained optimal machine learning model to predict antibody-antigen affinity;

[0019] Step S5: Affinity optimization of the antibody sequence is performed based on the differential evolution algorithm, and further filtering is performed by predicting the drugability of the antibody to screen out antibodies with high affinity and good drug properties.

[0020] Preferably, the language model includes one or more of ProtVec, ESM-1, ProteinBert, AntiBERTy, AntiBERTa and AbLang.

[0021] Preferably, the machine learning model includes Random Forest, XGBoost or Extra Trees.

[0022] Preferably, the step S5 adopts: using DE / best / 1 / bin differential evolution algorithm to optimize the affinity of the antibody sequence.

[0023] Preferably, the step S5 adopts:

[0024] Step S5.1: Initializing a set of initial antibody sequences through a differential evolution algorithm to establish an initial antibody library;

[0025] Step S5.2: Using the artificial intelligence model to evaluate each antibody sequence and calculate its affinity score with the target antigen;

[0026] Step S5.3: According to the fitness scores of the antibody sequences, antibody sequences that meet the preset requirements are selected as the next generation of the population;

[0027] Step S5.4: Generate new antibody sequences using crossover and mutation operations in the differential evolution algorithm to increase the diversity of the population;

[0028] Step S5.5: Replace or merge the newly generated antibody sequence with the original population to update the population; repeatedly trigger steps S5.3 to S5.5 until convergence or the maximum number of iterations is reached.

[0029] Preferably, the step S5 comprises: inputting the optimized antibody sequence into a drugability prediction model to predict the drugability characteristics of the antibody;

[0030] The drugability characteristics of the antibody include: affinity capture self-interaction nanoparticle spectroscopy, clonal self-interaction based on biolayer interferometry, multi-specific reagent binding, baculovirus particle ELISA, cross-interaction chromatography, enzyme-linked immunosorbent assay, HEK cell expression titer, hydrophobic interaction chromatography, salt gradient affinity capture self-interaction nanoparticle spectroscopy, vertical monolayer adsorption chromatography, size exclusion chromatography and differential scanning fluorescence quantification.

[0031] An antibody screening system based on a differential evolution algorithm and an artificial intelligence model provided by the present invention comprises:

[0032] Module M1: Obtaining antibody-antigen complex sequence and affinity data;

[0033] Module M2: Use the language model to encode the antibody and antigen sequences and convert them into matrix M;

[0034] Module M3: train the machine learning model using the matrix M and affinity parameters to obtain the optimal machine learning model after training;

[0035] Module M4: Predict antibody-antigen affinity using the trained optimal machine learning model;

[0036] Module M5: Affinity optimization of antibody sequences is performed based on the differential evolution algorithm, and further filtering is performed by predicting the drugability of antibodies to screen out antibodies with high affinity and good drug properties.

[0037] Preferably, the language model includes one or more of ProtVec, ESM-1, ProteinBert, AntiBERTy, AntiBERTa and AbLang.

[0038] Preferably, the machine learning model includes Random Forest, XGBoost or Extra Trees.

[0039] Preferably, the module M5 adopts: using DE / best / 1 / bin differential evolution algorithm to optimize the affinity of antibody sequence;

[0040] The module M5 adopts:

[0041] Module M5.1: Initialize a set of initial antibody sequences through differential evolution algorithm to establish an initial antibody library;

[0042] Module M5.2: Use the artificial intelligence model to evaluate each antibody sequence and calculate its affinity score with the target antigen;

[0043] Module M5.3: Based on the fitness scores of the antibody sequences, select the antibody sequences that meet the preset requirements as the next generation of the population;

[0044] Module M5.4: Generate new antibody sequences using crossover and mutation operations in differential evolution algorithms to increase population diversity;

[0045] Module M5.5: Replace or merge the newly generated antibody sequence with the original population to update the population; repeatedly trigger modules M5.3 to M5.5 until convergence or the maximum number of iterations is reached;

[0046] The module M5 adopts: inputting the optimized antibody sequence into the drugability prediction model to predict the drugability characteristics of the antibody;

[0047] The drugability characteristics of the antibody include: affinity capture self-interaction nanoparticle spectroscopy, clonal self-interaction based on biolayer interferometry, multi-specific reagent binding, baculovirus particle ELISA, cross-interaction chromatography, enzyme-linked immunosorbent assay, HEK cell expression titer, hydrophobic interaction chromatography, salt gradient affinity capture self-interaction nanoparticle spectroscopy, vertical monolayer adsorption chromatography, size exclusion chromatography and differential scanning fluorescence quantification.

[0048] Compared with the prior art, the present invention has the following beneficial effects:

[0049] 1. The differential evolution algorithm and artificial intelligence model are comprehensively applied. Through multiple iterations of optimization, the affinity between the antibody and the antigen is gradually improved to obtain an excellent antibody sequence with high affinity. This algorithm should be able to make full use of the global search capability of the differential evolution algorithm, and at the same time combine the artificial intelligence model to perform affinity evaluation to find a high-affinity antibody design scheme and meet multiple design goals and constraints. The present invention can improve the affinity of antibodies, reduce design cycles and costs, and is applicable to antibody design and optimization problems in the field of biomedicine.

[0050] 2. As a global search strategy, the differential evolution algorithm has strong global search capabilities. Through the differential evolution algorithm, a large number of differential antibody design schemes can be generated, and potential excellent schemes can be screened out. The artificial intelligence model is introduced as an affinity evaluation strategy to improve the accuracy and efficiency of the design scheme. The artificial intelligence model can evaluate the affinity of each candidate antibody by extracting useful features and rules from a large amount of data, and quickly screen out antibodies with higher affinity. The combination of the differential evolution algorithm and the artificial intelligence model enables the algorithm to efficiently find the best antibody design scheme in a large-scale search space, and has good scalability and adaptability.

[0051] 3. The input of the algorithm of the present invention only requires sequence information to design antibodies, without the need for complex crystal structure modeling and analysis steps, which simplifies the process of data acquisition and saves time and computing resources. In addition, it reduces the cost of research and design, making antibody design more economical and efficient, and is particularly suitable for large-scale antibody libraries and high-throughput design in research and application. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] Other features, objects and advantages of the present invention will become more apparent from the detailed description of non-limiting embodiments made with reference to the following drawings:

[0053] Figure 1 A schematic diagram of a flow chart of a method for training an antibody-antigen binding affinity model provided in an embodiment of the present invention.

[0054] Figure 2 A schematic diagram of the flow of an antibody design optimization algorithm combining a differential evolution algorithm and an artificial intelligence model provided in an embodiment of the present invention.

[0055] Figure 3 A basic structural diagram of an optional computer device provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0056] The present invention is described in detail below in conjunction with specific embodiments. The following embodiments will help those skilled in the art to further understand the present invention, but are not intended to limit the present invention in any form. It should be noted that, for those of ordinary skill in the art, several changes and improvements can also be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

[0057] Example 1

[0058] In order to overcome the limitations of traditional methods and existing models, this paper proposes an antibody design optimization algorithm that combines differential evolution algorithm and artificial intelligence model. The algorithm aims to improve the efficiency and accuracy of antibody design to meet the demand for high-affinity antibodies in the biopharmaceutical field.

[0059] According to the present invention, there is provided an antibody screening method combining a differential evolution algorithm and an artificial intelligence model, comprising:

[0060] Step S1: the acquisition module acquires the data of the sequence and affinity of the antibody-antigen complex;

[0061] Step S2: The processing module encodes the antibody and antigen sequences using a language model to represent their transformation matrix M;

[0062] Step S3: Input the matrix M and affinity parameters into the machine learning model, adjust the parameters of the model, and when the loss value between the predicted sample affinity parameters and the corresponding actual affinity parameters meets the preset loss condition, end the training, and determine the machine learning model at the end of the training as the optimal model for subsequent antibody design.

[0063] Step S4: predicting antibody-antigen affinity through the trained optimal machine learning model;

[0064] Step S5: Affinity optimization of the antibody sequence is performed based on the differential evolution algorithm, and further filtering is performed by predicting the drugability of the antibody to screen out antibodies with high affinity and good drug properties.

[0065] The present invention uses a language model to encode antibody and antigen sequences, and applies a machine learning algorithm to predict. The method of the present invention can accurately predict the affinity of antibody antigens. By adjusting model parameters and using technical means such as cross-validation, a prediction result with high accuracy can be obtained. At the same time, the present invention optimizes the affinity of antibody sequences and screens the drugability based on the differential evolution algorithm, and the method of the present invention can recommend a group of optimized antibody sequences. These sequences are adjusted and optimized to have a stronger ability to bind to the target antigen and have good biophysical properties, providing valuable recommended sequences for antibody design and treatment plans.

[0066] Specifically, the step S1 adopts: the antibody-antigen complex sequence and affinity data set can be obtained by database query, experimental measurement and the like.

[0067] Specifically, the step S2 adopts a pre-trained language model, such as ProtVec, AntiBERTy and AbLang, to encode the antibody and antigen sequences to capture their characteristics and interaction information.

[0068] Specifically, the step S3 may use a variety of machine learning algorithms, such as support vector machines, random forests, etc., to train a prediction model for antibody-antigen affinity to evaluate the affinity of the antibody design solution.

[0069] Specifically, in step S5, any one or several CDR regions of the heavy and light chains of the antibody can be selected for design.

[0070] The step S5 uses a differential evolution algorithm to initialize a group of initial antibody sequences to establish an initial antibody library.

[0071] When calculating fitness, an artificial intelligence model is used to evaluate each antibody sequence and calculate its affinity score with the target antigen.

[0072] According to the fitness scores of the antibody sequences, a selection operation is used to select a portion of excellent antibody sequences as the next generation of the population.

[0073] The crossover and mutation operations in the differential evolution algorithm are used to generate new antibody sequences to increase the diversity of the population.

[0074] The newly generated antibody sequences are replaced or merged with the original population to update the population.

[0075] According to the set number of iterations or termination conditions, selection, mutation, crossover and optimization operations are continuously performed to gradually optimize the antibody design scheme until convergence or the maximum number of iterations is reached.

[0076] The step S5 adopts: antibody drugability prediction including AC-SINS (affinity capture self-interaction nanoparticle spectroscopy), CSIBLI (clonal self-interaction based on biolayer interferometry), PSR (multi-specific reagent binding), BVP-ELISA (baculovirus particle ELISA), CIC (cross-interaction chromatography), ELISA (enzyme-linked immunosorbent assay), HEK (HEK cell expression titer), HIC (hydrophobic interaction chromatography), SGAC-SINS (salt gradient affinity capture self-interaction nanoparticle spectroscopy), SMAC (vertical monolayer adsorption chromatography), SEC (size exclusion chromatography), DSF (differential scanning fluorescence quantification).

[0077] According to a computer program product or computer program provided by the present invention, the computer program product or computer program includes computer instructions, which are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above antibody design optimization method based on the differential evolution algorithm and the artificial intelligence model.

[0078] According to a storage medium storing computer-readable instructions provided by the present invention, when the computer-readable instructions are executed by one or more processors, the one or more processors execute the steps of the antibody design optimization method based on the differential evolution algorithm and the artificial intelligence model as described above.

[0079] A computer device provided according to the present invention includes a memory and a processor, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the processor executes the steps of the antibody design optimization method based on the differential evolution algorithm and the artificial intelligence model as described above.

[0080] The antibody screening system based on the differential evolution algorithm and the artificial intelligence model provided by the present invention can be implemented by the step flow in the antibody screening method based on the differential evolution algorithm and the artificial intelligence model provided by the present invention. Those skilled in the art can understand the antibody screening method based on the differential evolution algorithm and the artificial intelligence model as a preferred example of the antibody screening system based on the differential evolution algorithm and the artificial intelligence model.

[0081] Example 2

[0082] Embodiment 2 is a preferred embodiment of Embodiment 1

[0083] like Figure 2 As shown, an antibody screening system combining a differential evolution algorithm and an artificial intelligence model provided by the present invention includes:

[0084] Acquisition module: Acquisition of antibody heavy and light chain sequences or variable region sequences input by the user.

[0085] Processing module: Use language models to encode antibody and antigen sequences. You can choose to use one or more language models such as ProtVec, ESM-1, ProteinBert, AntiBERTy, AntiBERTa, and AbLang to convert antibody and antigen sequences into vector representations to capture the semantic and structural information of the sequences.

[0086] Training module: Figure 1 As shown, in the present invention, a machine learning model is used to train antibody and antigen data to predict antibody-antigen affinity. The specific steps of the training module are as follows:

[0087] Module M3.1: Dataset preparation: Obtain antibody-antigen sequence and affinity data from the CSM-AB dataset and online public data to construct training and test sets. The training set is used for model training, and the test set is used for model evaluation.

[0088] Module M3.2: Feature encoding: Use one or more language models such as ProtVec, ESM-1, ProteinBert, AntiBERTy, AntiBERTa, and AbLang to encode antibody and antigen sequences and convert them into feature vector representations. These language models are pre-trained based on large-scale sequence data and can capture key feature information in the sequence.

[0089] Module M3.3: Model selection and training: Select machine learning algorithms such as random forest, XGBoost or ExtraTrees, and use the encoded antibody and antigen feature vectors as input to train the model. The optimal model is obtained by adjusting model parameters, cross-validation and other methods. In this embodiment, the random forest model based on ProtVec encoding shows the best performance.

[0090] Module M3.4: Model evaluation: Use the test set to evaluate the trained model and calculate the error between the predicted result and the actual affinity. The evaluation indicators used include root mean square error (RMSE) and Pearson Correlation Coefficient.

[0091] Prediction module: Use the trained optimal machine learning model and the encoded antibody-antigen sequence input by the user to predict the antibody-antigen affinity and evaluate the effect of antibody design.

[0092] Optimization module: The differential evolution algorithm is used to optimize the antibody sequence. The differential evolution algorithm is a global optimization algorithm that searches for the optimal solution of the antibody sequence by simulating operations such as mutation, crossover and selection in the biological evolution process.

[0093] Using the DE / best / 1 / bin differential evolution algorithm, in this example, affinity maturation is performed on the CDR-H3 of the antibody. The CDR-H3 length is set to N, the population size is 192, and the maximum number of iterations is 150. F is a parameter used to control the mutation operation, which controls the scaling factor of the mutation vector in the differential calculation. F is set to 0.5. Cr is a parameter used to control the crossover operation. The crossover operation obtains a new solution by mixing the mutation vector with a parent individual. Cr is used to control the mixing ratio between the mutation vector and the parent individual. Cr is set to 0.7. The specific steps are as follows:

[0094] Module M5.1: Initialize the population: For i = 1, 2, ..., N, randomly generate an N-dimensional vector x iThat is, the CHR-H3 sequence is used as the initial population and encoded into binary using Real Integer Hybrid Encoding (RI).

[0095] P={x1,x2,…,x N},x i =(x i1 ,x i2 ,…,x i17 ) (1)

[0096] Module M5.2: Select the best individual: When calculating the fitness, use the best machine learning model to predict the affinity of the antibody binding to the antigen, and select the CDR-H3 sequence with the highest affinity from the initial population as the optimal sequence.

[0097]

[0098] Among them, f(x i ) represents the fitness function, that is, we construct the optimal model for predicting affinity.

[0099] Module M5.3: Mutation operation: For each antibody sequence, for each individual, two different individuals r1 and r2 are randomly selected from the population, the difference vector between them is calculated, and then the difference vector is weighted and synthesized to obtain the mutant individual v i The differential mutation operation utilizes a differential vector to change a portion of the antibody sequence to introduce diversity and explore new solution spaces.

[0100] v i =x best +F*(r1-r2) (3)

[0101] Module M5.4: Crossover operation: Use binary crossover to cross the variant individual with the original antibody sequence to generate a new individual u i For each dimension d, generate a random number j, if j is less than or equal to the crossover factor Cr or j is equal to a random number J rand , then use the mutant individual v i The corresponding element v i,d Replace the original individual x i The corresponding element x i,d .

[0102] The crossover operation combines partial genetic information of two antibody sequences to produce an offspring sequence with affinity advantages.

[0103]

[0104] Among them, j is the number generated by the random function rand, d is the dimension of the individual, Cr is the cross factor, and J rand Dimensions randomly chosen by the rand function.

[0105] Module M5.5: Selection operation: According to the affinity predicted by the optimal model, antibody sequences with higher affinity are selected as the next generation population. The selection operation retains individuals with higher fitness and promotes the transmission and accumulation of excellent solutions.

[0106]

[0107] Module M5.6: Iterative optimization: Repeat modules M5.2 to 5.5 until the stopping condition is reached (e.g., reaching the maximum number of iterations or converging to the optimal solution).

[0108] The optimized antibody sequence is input into the drugability prediction model Abpred to predict the drugability characteristics of the antibody. These indicators include affinity capture self-interaction nanoparticle spectroscopy (AC-SINS), clone self-interaction using bio-layer interferometry technology (CSI-BLI) and salt gradient affinity capture self-interaction nanoparticle spectroscopy (SGAC-SINS) to evaluate antibody self-interaction; flow cytometry of polyspecificity reagent (PSR) and enzyme-linked immunosorbent assay (ELISA) to evaluate antibody specificity; cross-interaction chromatography (CIC) to determine antibody solubility; baculovirus particle assay-enzyme-linked immunosorbent assay (Baculovirus Particle assay-enzyme-linked immunosorbent The antibody serum clearance rate was determined by BVP-ELISA assay; the thermal stability of the antibody was determined by Differential Scanning Fluorimetry (DSF); the antibody expression level was determined by HEK cell expression titer; the colloidal stability of the antibody was determined by Standup Monolayer Adsorption Chromatography (SMAC) and the accelerated stability (AS) of the antibody was determined by Size-Exclusion Chromatography (SEC). Based on the prediction results, the antibody design scheme with good drugability was screened out.

[0109] The present invention proposes an antibody affinity optimization algorithm that combines a differential evolution algorithm and an artificial intelligence model. The algorithm only requires antibody antigen sequence information to perform antibody design, without the need for complex crystal structure modeling and analysis steps, which simplifies the data acquisition process and saves time, computing resources and research costs. Using the differential evolution algorithm to perform a global search, a large number of antibody design schemes with differences can be generated. By introducing an artificial intelligence model to evaluate the affinity of candidate antibodies, antibodies with higher affinity can be quickly screened out. The combination of the differential evolution algorithm and the artificial intelligence model enables the algorithm to efficiently find the best antibody design scheme in a large-scale search space, and has good scalability and adaptability.

[0110] like Figure 3 As shown, in the present invention, in order to implement the execution of a computer program product or a computer program, it is necessary to adapt corresponding hardware devices. These hardware devices may include computers, servers, processors, memory and storage devices, etc. The processor of the computer device reads the computer instructions in the storage medium and executes the computer instructions, so that the computer device can execute the antibody design optimization algorithm of the present invention.

[0111] In the implementation process of the present invention, appropriate hardware devices can be selected to support the execution of the algorithm according to actual needs and computing resources. Reasonable hardware configuration can improve the operating efficiency and performance of the algorithm, speed up the optimization process, and further enhance the effect of antibody design. For example, a high-performance computer or a server with sufficient computing power and storage capacity can be used. The computer or server should be equipped with hardware components such as appropriate processors, memory, storage devices, and network connections to ensure that the algorithm can be effectively executed and large-scale data sets can be processed. In addition, it is also possible to consider using an accelerator (such as a GPU) to speed up the calculation speed.

[0112] Those skilled in the art know that, in addition to implementing the system, device and its various modules provided by the present invention in a purely computer-readable program code, it is entirely possible to implement the same program in the form of logic gates, switches, application-specific integrated circuits, programmable logic controllers and embedded microcontrollers by logically programming the method steps. Therefore, the system, device and its various modules provided by the present invention can be considered as a hardware component, and the modules included therein for implementing various programs can also be considered as structures within the hardware component; the modules for implementing various functions can also be considered as both software programs for implementing the method and structures within the hardware component.

[0113] The above describes the specific embodiments of the present invention. It should be understood that the present invention is not limited to the above specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which does not affect the essence of the present invention. In the absence of conflict, the embodiments of the present application and the features in the embodiments can be combined with each other arbitrarily.

Claims

1. An antibody screening method based on differential evolution algorithm and artificial intelligence model, characterized in that: include: Step S1: obtaining antibody-antigen complex sequence and affinity data; Step S2: Encode the antibody and antigen sequences using a language model and convert them into a matrix M; Step S3: using the matrix M and affinity parameters to train the machine learning model to obtain the optimal machine learning model after training; Step S4: using the trained optimal machine learning model to predict antibody-antigen affinity; Step S5: Affinity optimization of the antibody sequence is performed based on the differential evolution algorithm, and further filtering is performed by predicting the drugability of the antibody to screen out antibodies with high affinity and good drug properties.

2. The antibody screening method based on differential evolution algorithm and artificial intelligence model according to claim 1, characterized in that: The language model includes one or more of ProtVec, ESM-1, ProteinBert, AntiBERTy, AntiBERTa and AbLang.

3. The antibody screening method based on differential evolution algorithm and artificial intelligence model according to claim 1, characterized in that: The machine learning model includes random forest, XGBoost or Extra Trees.

4. The antibody screening method based on differential evolution algorithm and artificial intelligence model according to claim 1, characterized in that: The step S5 adopts: using DE / best / 1 / bin differential evolution algorithm to optimize the affinity of the antibody sequence.

5. The antibody screening method based on differential evolution algorithm and artificial intelligence model according to claim 1, characterized in that: The step S5 adopts: Step S5.1: Initializing a set of initial antibody sequences through a differential evolution algorithm to establish an initial antibody library; Step S5.2: Using the artificial intelligence model to evaluate each antibody sequence and calculate its affinity score with the target antigen; Step S5.3: According to the fitness scores of the antibody sequences, antibody sequences that meet the preset requirements are selected as the next generation of the population; Step S5.4: Generate new antibody sequences using crossover and mutation operations in the differential evolution algorithm to increase the diversity of the population; Step S5.5: Replace or merge the newly generated antibody sequence with the original population to update the population; Steps S5.3 to S5.5 are triggered repeatedly until convergence or the maximum number of iterations is reached.

6. The antibody screening method based on differential evolution algorithm and artificial intelligence model according to claim 1, characterized in that: The step S5 comprises: inputting the optimized antibody sequence into a drugability prediction model to predict the drugability characteristics of the antibody; The drugability characteristics of the antibody include: affinity capture self-interaction nanoparticle spectroscopy, clonal self-interaction based on biolayer interferometry, multi-specific reagent binding, baculovirus particle ELISA, cross-interaction chromatography, enzyme-linked immunosorbent assay, HEK cell expression titer, hydrophobic interaction chromatography, salt gradient affinity capture self-interaction nanoparticle spectroscopy, vertical monolayer adsorption chromatography, size exclusion chromatography and differential scanning fluorescence quantification.

7. An antibody screening system based on differential evolution algorithm and artificial intelligence model, characterized in that: include: Module M1: Obtaining antibody-antigen complex sequence and affinity data; Module M2: Use the language model to encode the antibody and antigen sequences and convert them into matrix M; Module M3: train the machine learning model using the matrix M and affinity parameters to obtain the optimal machine learning model after training; Module M4: Predict antibody-antigen affinity using the trained optimal machine learning model; Module M5: Affinity optimization of antibody sequences is performed based on the differential evolution algorithm, and further filtering is performed by predicting the drugability of antibodies to screen out antibodies with high affinity and good drug properties.

8. The antibody screening system based on differential evolution algorithm and artificial intelligence model according to claim 7, characterized in that: The language model includes one or more of ProtVec, ESM-1, ProteinBert, AntiBERTy, AntiBERTa and AbLang.

9. The antibody screening system based on differential evolution algorithm and artificial intelligence model according to claim 7, characterized in that: The machine learning model includes random forest, XGBoost or Extra Trees.

10. The antibody screening system based on differential evolution algorithm and artificial intelligence model according to claim 7, characterized in that: The module M5 adopts: using DE / best / 1 / bin differential evolution algorithm to optimize the affinity of antibody sequence; The module M5 adopts: Module M5.1: Initialize a set of initial antibody sequences through differential evolution algorithm to establish an initial antibody library; Module M5.2: Use the artificial intelligence model to evaluate each antibody sequence and calculate its affinity score with the target antigen; Module M5.3: Based on the fitness scores of the antibody sequences, select the antibody sequences that meet the preset requirements as the next generation of the population; Module M5.4: Generate new antibody sequences using crossover and mutation operations in differential evolution algorithms to increase population diversity; Module M5.5: Replace or merge the newly generated antibody sequence with the original population to update the population; repeatedly trigger modules M5.3 to M5.5 until convergence or the maximum number of iterations is reached; The module M5 adopts: inputting the optimized antibody sequence into the drugability prediction model to predict the drugability characteristics of the antibody; The drugability characteristics of the antibody include: affinity capture self-interaction nanoparticle spectroscopy, clonal self-interaction based on biolayer interferometry, multi-specific reagent binding, baculovirus particle ELISA, cross-interaction chromatography, enzyme-linked immunosorbent assay, HEK cell expression titer, hydrophobic interaction chromatography, salt gradient affinity capture self-interaction nanoparticle spectroscopy, vertical monolayer adsorption chromatography, size exclusion chromatography and differential scanning fluorescence quantification.

Images