An agent-based full-process antibody drug design optimization system and method

Abstract

The application provides an intelligent agent-based whole-process antibody drug design optimization system and method, which comprises the following steps: receiving an antibody design requirement input by a user through an intelligent agent, performing intention recognition, task classification and parameter extraction on the input by the intelligent agent, and automatically constructing an antibody design task plan; structural modeling and combination hotspot analysis; generating a brand new antibody skeleton or optimizing an existing antibody skeleton; CDR sequence filling and candidate sequence generation; structure verification and affinity evaluation; humanization design and developability evaluation; candidate antibody sorting and result output; and generating a comprehensive design report. The application is used for guiding a user to complete a complete process from a target point to a high-affinity humanized antibody candidate sequence.

Description

Technical Field

[0001] This invention relates to the field of antibody drug research technology, and more specifically, to a full-process antibody drug design and optimization system and method based on intelligent agents. Background Technology

[0002] Antibody drugs, due to their high specificity and low toxicity, have become an important research and development direction in the biopharmaceutical field. Traditional antibody discovery methods (such as hybridoma technology and phage display technology) suffer from problems such as long development cycles, high costs, and low screening efficiency. In recent years, with the development of deep learning technology, computation-assisted antibody design has gradually emerged. Although there are currently various independent tools for protein structure prediction, molecular docking, or sequence generation, existing technologies still have the following significant drawbacks:

[0003] 1. Fragmented tools: Existing software usually exists as a single tool (e.g., only for modeling or only for humanization), and the data formats between tools are not interoperable, lacking a unified process to manage the data throughout the entire process.

[0004] 2. High operational threshold: Most tools require users to have a very high background in programming or structural biology, require manual configuration of a large number of parameters, and lack intelligent guidance.

[0005] 3. Lack of context awareness: Existing tools are unable to automatically adjust the next step's strategy based on the calculation results of the previous step (such as antigen hotspot analysis), and cannot form a closed-loop optimization.

[0006] 4. Difficulty in balancing humanization and affinity: Traditional humanization modification often leads to a significant decrease in affinity, and there is a lack of an integrated multi-objective optimization scheme that can simultaneously balance the reduction of immunogenicity and the maintenance of affinity.

[0007] Therefore, there is a need for an agent-based end-to-end antibody design and optimization system and method that can guide users through the entire process from target to high-affinity humanized antibody candidate sequence. Summary of the Invention

[0008] To address the shortcomings of existing technologies, the present invention aims to provide an agent-based end-to-end antibody drug design and optimization system and method, which guides users through the entire process from target to high-affinity humanized antibody candidate sequence.

[0009] To solve the above problems, the technical solution of the present invention is as follows:

[0010] A full-process antibody drug design and optimization system based on intelligent agents, comprising:

[0011] The intelligent interactive agent module is used to receive antibody design requirements input by users in the form of natural language, sequence files or structure files, and to identify user intentions, allocate tasks and extract parameters through an intelligent agent, and transform the user's unstructured design goals into executable antibody design constraints.

[0012] The task planning and status management module is used to automatically construct the antibody design workflow according to the antibody design constraints, manage the execution status of each agent, task queue, user confirmation information, intermediate results and tool call results, and dynamically adjust the task plan according to the results of structure prediction, sequence generation, humanization evaluation or affinity prediction.

[0013] The structural modeling and analysis module is used to call the protein structure prediction model to perform three-dimensional structural modeling on the input antigen sequence, antibody sequence, or antigen-antibody complex. It further analyzes the antigen epitope, antibody CDR region, antigen-antibody binding interface, spatial proximity, structural confidence, steric hindrance conflict, and binding hotspots, providing structural constraints for subsequent antibody backbone generation and sequence optimization.

[0014] The de novo sequence generation module includes a backbone generation model and a sequence filling model. The backbone generation model generates a candidate antibody backbone structure that is spatially complementary to the antigen surface based on the antigen's three-dimensional structure, target epitopes or binding hotspots, and a preset CDR design region. The sequence filling model performs amino acid sequence filling, redesign, and optimization on the CDR region or a specified design region based on the candidate antibody backbone structure, antigen-antibody interface geometry, and local amino acid environment to generate multiple candidate antibody sequences.

[0015] The structure validation and candidate screening module is used to re-predict the structure of antigen-antibody complexes from the generated candidate antibody sequences, and to screen and rank candidate antibodies based on CDR conformational retention, epitope contact, interface shape complementarity, structural confidence, spatial conflict, interface interaction and candidate sequence diversity.

[0016] The sequence optimization and evaluation module includes a humanization model and a feasibility evaluation model. The humanization model performs homology searches, numbering, alignment, and frame region analysis on candidate antibody sequences against a human antibody germline gene database. It identifies the V-gene and J-gene with the highest similarity or best overall score as humanization reference templates and designs humanization mutations in frame regions or low-risk sites while preserving the conformation of the CDR region and key binding sites. The feasibility evaluation model assesses the immunogenicity risk, hydrophobicity, aggregation tendency, charge distribution, stability, expression feasibility, and sequence abnormality characteristics of candidate antibody sequences.

[0017] The affinity prediction and mutation impact assessment module is used to call the protein-protein binding affinity prediction model to predict the binding free energy or affinity score of antibody-antigen complexes before and after humanization, compare the impact of humanization mutations, CDR redesign mutations or frame region mutations on binding ability, and combine structural stability, interfacial interaction and exploitability indicators to comprehensively rank the final candidate antibody sequences.

[0018] Preferably, the antibody design constraints include one or more of the following: target antigen, antibody type, target epitope or binding hotspot, CDR region to be optimized, CDR length range, affinity optimization target, humanization target, and exploitability requirements.

[0019] Preferably, the system further includes a report generation module, which automatically summarizes user input information, task plans, structural modeling results, epitope and hotspot analysis results, candidate antibody backbones, candidate antibody sequences, humanization schemes, affinity prediction results, exploitability assessment results, and recommended experimental verification schemes to generate a standardized antibody design optimization report.

[0020] Furthermore, the present invention also provides a full-process antibody drug design and optimization method based on intelligent agents, comprising the following steps:

[0021] The intelligent agent receives the antibody design requirements input by the user, performs intent recognition, task classification and parameter extraction on the input, and automatically constructs an antibody design task plan.

[0022] Structural modeling combined with hotspot analysis;

[0023] Generate entirely new antibody backbones or optimize existing antibody backbones;

[0024] CDR sequence filling and candidate sequence generation;

[0025] Structural validation and affinity assessment;

[0026] Human-centered design and development potential assessment;

[0027] Candidate antibody ranking and result output;

[0028] Generate a comprehensive design report.

[0029] Preferably, the structural modeling and hotspot analysis steps specifically include: performing three-dimensional structural prediction on the input antigen sequence, antibody sequence, or antigen-antibody complex to obtain the antigen structure, antibody structure, or antigen-antibody complex structure; and analyzing the antigen epitopes, binding hotspots, CDR regions, antigen-antibody interfaces, spatial distances, steric hindrance conflicts, and structural confidence based on the structure to determine the structural constraints for subsequent antibody design.

[0030] Preferably, the step of generating a novel antibody backbone or optimizing an existing antibody backbone specifically includes: generating a novel antibody backbone structure based on the target antigen structure, binding epitopes or hotspot residues, and a preset CDR design region, or redesigning one or more CDR regions in an existing antibody backbone to obtain a set of candidate antibody backbones that are spatially complementary to the target antigen.

[0031] Preferably, the steps of CDR sequence filling and candidate sequence generation specifically include: filling or optimizing the amino acid sequence of a specified CDR region based on the candidate antibody backbone structure, antigen-antibody interface geometric features and local structural environment to generate multiple candidate antibody sequences, and performing preliminary screening based on sequence rationality, structural compatibility, interface contact features and diversity.

[0032] Preferably, the steps of structure verification and affinity assessment specifically include: predicting the complex structure of the candidate antibody sequence and the target antigen, assessing whether the candidate sequence maintains the expected CDR conformation, whether it forms effective contact with the target epitope, and whether there is significant spatial conflict; further predicting the binding free energy or affinity score of the candidate antibody and the antigen, and assessing the influence of different candidate sequences and mutation sites on the binding ability.

[0033] Preferably, the steps of humanization design and developmentability assessment specifically include: performing homology search and comparison between the candidate antibody sequence and the human antibody germline gene database to determine the humanization reference template; designing humanization mutations for the candidate antibody while preserving the CDR region, key binding residues and antigen-binding conformation, and assessing the changes in affinity, immunogenicity risk, stability, hydrophobicity, aggregation tendency, expression feasibility and overall developmentability of the sequence before and after humanization.

[0034] Preferably, the steps of ranking and outputting candidate antibodies specifically include: ranking the candidate antibodies comprehensively based on structural quality, affinity prediction results, humanization degree, exploitability score, sequence diversity, and user-defined optimization goals to form a recommended candidate antibody set.

[0035] Compared with the prior art, the present invention has the following beneficial effects:

[0036] 1. Intelligent integration of the entire process: This invention innovatively introduces a large language model into the antibody design process, linking antigen modeling, backbone generation, sequence filling, and humanization modification. The intelligent agent can select to start from any link based on the user's intention to design antibody design tasks. This solves the problem of fragmented tools in current antibody design and optimization, integrates tools related to antibody design into a platform, and improves R&D efficiency.

[0037] 2. Lowering the barrier to entry: With the guidance of intelligent agents, users only need to describe their design intentions in simple natural language, and the system can automatically plan the best computational path, enabling biologists without a computational background to perform complex antibody design.

[0038] 3. Dynamic Feedback for Design Modification: Supports real-time feedback based on antibody-antigen binding modes, enabling fine-grained redesign of CDR regions for specific binding hotspots, thus improving the positive rate of designed antibodies. Human experts can conduct multiple design attempts on binding hotspots of interest to obtain various antibody structures binding to different antigens. Attached Figure Description

[0039] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0040] Figure 1 This is a block diagram of the full-process antibody drug design and optimization system based on intelligent agents according to the present invention;

[0041] Figure 2 This is a flowchart of the full-process antibody drug design and optimization method based on intelligent agents according to the present invention. Detailed Implementation

[0042] The present invention will now be described in detail with reference to specific embodiments. These embodiments will help those skilled in the art to further understand the present invention, but do not limit the invention in any way. It should be noted that those skilled in the art can make several changes and improvements without departing from the concept of the present invention. These all fall within the protection scope of the present invention.

[0043] Antibody drugs are a core treatment for many diseases, but traditional antibody drug design and optimization processes suffer from numerous bottlenecks, including low automation, fragmented processes leading to data silos, prominent contradictions in multi-objective optimization making it difficult to balance the overall performance of antibodies, rigid interaction methods, and existing AI methods focusing only on a single step and lacking global coordination. This results in low R&D efficiency, high costs, and slow clinical translation. To address these core pain points, this invention proposes a full-process antibody drug design and optimization system and method based on intelligent agents.

[0044] Specifically, such as Figure 1 As shown, the intelligent agent-based full-process antibody drug design and optimization system of the present invention includes an intelligent interactive agent module 1, a task planning and state management module 2, a structure modeling and analysis module 3, a de novo sequence generation module 4, a structure verification and candidate screening module 5, a sequence optimization evaluation module 6, an affinity prediction and mutation impact evaluation module 7, and a report generation module 8.

[0045] The intelligent interactive agent module serves as the user's interface with the system. It receives antibody design requirements input by the user in the form of natural language, sequence files, or structured files. Through an intelligent agent, it identifies the user's intent, performs task allocation, and extracts parameters, transforming the user's unstructured design goals into executable antibody design constraints. These antibody design constraints include one or more of the following: target antigen, antibody type, target epitope or binding hotspot, CDR region to be optimized, CDR length range, affinity optimization target, humanization target, and exploitability requirements.

[0046] The task planning and status management module is used to automatically construct the antibody design workflow according to the design constraints, manage the execution status of each agent, task queue, user confirmation information, intermediate results and tool call results, and dynamically adjust the task plan according to the results of structure prediction, sequence generation, humanization evaluation or affinity prediction, so as to realize the traceability, reproducibility and closed-loop iteration of the antibody design process.

[0047] The structure modeling and analysis module is used to call the protein structure prediction model to perform three-dimensional structure modeling on the input antigen sequence, antibody sequence, or antigen-antibody complex. It further analyzes the antigen epitope, antibody CDR region, antigen-antibody binding interface, spatial proximity, structural confidence, steric hindrance conflict, and binding hotspots, providing structural constraints for subsequent antibody backbone generation and sequence optimization.

[0048] The de novo sequence generation module includes a backbone generation model and a sequence filling model. The backbone generation model generates a candidate antibody backbone structure that is spatially complementary to the antigen surface based on the antigen's three-dimensional structure, target epitopes or binding hotspots, and a preset CDR design region. The sequence filling model performs amino acid sequence filling, redesign, and optimization on the CDR region or a specified design region based on the candidate antibody backbone structure, antigen-antibody interface geometry, and local amino acid environment to generate multiple candidate antibody sequences.

[0049] The structure verification and candidate screening module is used to re-predict the antigen-antibody complex structure of the generated candidate antibody sequences, and to screen and rank the candidate antibodies based on CDR conformational retention, epitope contact, interface shape complementarity, structural confidence, spatial conflict, interface interaction and candidate sequence diversity.

[0050] The sequence optimization evaluation module includes a humanization model and a feasibility evaluation model. The humanization model performs homology searches, numbering, alignment, and frame region analysis on candidate antibody sequences against a human antibody germline gene database. It identifies the V-gene and J-gene with the highest similarity or best overall score as humanization reference templates and designs humanization mutations in frame regions or low-risk sites while preserving the conformation of the CDR region and key binding sites. The feasibility evaluation model assesses the immunogenicity risk, hydrophobicity, aggregation tendency, charge distribution, stability, expression feasibility, and sequence abnormality characteristics of candidate antibody sequences.

[0051] The affinity prediction and mutation impact assessment module is used to call the protein-protein binding affinity prediction model to predict the binding free energy or affinity score of antibody-antigen complexes before and after humanization, compare the impact of humanization mutations, CDR redesign mutations or frame region mutations on binding ability, and combine structural stability, interfacial interaction and exploitability indicators to comprehensively rank the final candidate antibody sequences.

[0052] The report generation module is used to automatically summarize user input information, task plans, structural modeling results, epitope and hotspot analysis results, candidate antibody backbones, candidate antibody sequences, humanization schemes, affinity prediction results, developmentability assessment results, and recommended experimental verification schemes to generate a standardized antibody design optimization report.

[0053] Furthermore, this invention also provides a full-process antibody drug design and optimization method based on intelligent agents, such as... Figure 2 As shown, the method includes the following steps:

[0054] S1: The intelligent agent receives the antibody design requirements input by the user. The intelligent agent performs intent recognition, task classification and parameter extraction on the input, and automatically constructs an antibody design task plan.

[0055] Specifically, the design requirements analysis and task planning involves receiving antibody design requirements input by the user through an intelligent agent. The input includes one or more of the following: natural language description, antigen sequence, antigen structure, known antibody sequence, known antibody structure, target epitope information, or design constraint parameters. The intelligent agent performs intent recognition, task classification, and parameter extraction on the input and automatically constructs an antibody design task plan.

[0056] S2: Structural modeling combined with hotspot analysis;

[0057] Specifically, the three-dimensional structure of the input antigen sequence, antibody sequence, or antigen-antibody complex is predicted to obtain the antigen structure, antibody structure, or antigen-antibody complex structure. Based on the structure, the antigen epitopes, binding hotspots, CDR regions, antigen-antibody interfaces, spatial distances, steric conflicts, and structural confidence are analyzed to determine the structural constraints for subsequent antibody design.

[0058] S3: Generate entirely new antibody backbones or optimize existing antibody backbones;

[0059] Specifically, based on the target antigen structure, binding epitopes or hotspot residues, and pre-defined CDR design regions, novel antibody backbone structures are generated, or one or more CDR regions in an existing antibody backbone are redesigned to obtain a set of candidate antibody backbones that are spatially complementary to the target antigen.

[0060] S4: CDR sequence filling and candidate sequence generation;

[0061] Specifically, based on the candidate antibody backbone structure, antigen-antibody interface geometry, and local structural environment, amino acid sequence filling or sequence optimization is performed on the specified CDR region to generate multiple candidate antibody sequences, and preliminary screening is conducted based on sequence rationality, structural compatibility, interface contact characteristics, and diversity.

[0062] S5: Structural verification and affinity assessment;

[0063] Specifically, the complex structure of candidate antibody sequences and target antigens is predicted to assess whether the candidate sequences maintain the expected CDR conformation, whether they form effective contact with the target epitope, and whether there is significant spatial conflict. Furthermore, the binding free energy or affinity score of candidate antibodies and antigens is predicted to assess the impact of different candidate sequences and mutation sites on binding ability.

[0064] S6: Human-centered design and development potential assessment;

[0065] Specifically, the candidate antibody sequences are homology-searched and compared with the human antibody germline gene database to determine the humanization reference template; while preserving the CDR region, key binding residues and antigen-binding conformation, humanization mutations are designed for the candidate antibodies, and the changes in affinity, immunogenicity risk, stability, hydrophobicity, aggregation tendency, expression feasibility and overall development potential of the sequences before and after humanization are evaluated.

[0066] S7: Candidate antibody ranking and result output;

[0067] Specifically, candidate antibodies are comprehensively ranked based on structural quality, affinity prediction results, humanization degree, exploitability score, sequence diversity, and user-defined optimization goals to form a recommended candidate antibody set.

[0068] S8: Generate a comprehensive design report.

[0069] Specifically, the antibody design process, user input parameters, task execution path, antigen and antibody structural information, epitope and hotspot analysis results, candidate antibody sequences, structural prediction results, humanization schemes, affinity prediction results, developmentability assessment results, and experimental validation suggestions are summarized to generate a comprehensive design report.

[0070] The following specific embodiments illustrate the agent-based full-process antibody drug design and optimization method of the present invention.

[0071] Example 1: De novo antibody design based on antigen sequence

[0072] This example demonstrates the process by which a user provides a new target antigen and the system automatically generates an antibody. The specific steps are as follows:

[0073] Input and Modeling: The user inputs the FASTA sequence of the antigen into the intelligent agent interface, and the system calls the structural modeling module to generate the 3D structure of the antigen;

[0074] Hotspot definition: The agent guides the user to select binding hotspots on the antigen surface in a visual interface;

[0075] Antibody backbone generation: The system generates an antibody variable region backbone that binds to the selected antigen hotspots on the antigen surface.

[0076] CDR sequence filling: Keeping the backbone structure unchanged, the system uses a deep learning algorithm to fill the amino acid sequence of the antibody complementarity-determining region (CDR) to generate multiple initial antibody sequences;

[0077] Humanization and optimization: The system automatically identifies regions in the generated sequence that differ significantly from human germline genes, and performs residue substitution while ensuring affinity to complete humanization;

[0078] Output: The system output includes the CDR sequence, the predicted 3D complex structure, and a report.

[0079] Example 2: Antibody CDR Region Redesign

[0080] This embodiment is applicable to scenarios where a user already has an antibody sequence but wishes to improve its affinity or change its binding properties. The specific steps are as follows:

[0081] Sequence import: Users only need to upload existing antibody sequences;

[0082] Structure recovery: The system constructs the current antibody CDR structural region through homology modeling;

[0083] Region locking and masking: When the user specifies the region to be optimized (e.g., only optimize CDR-H3), the system will lock other regions (FR region, CDR-L, etc.);

[0084] Sequence sampling: The generative design module performs high-throughput virtual mutation and sampling on the specified CDR region while preserving the main structure of the antibody, to find other sequence combinations with lower binding free energy or the same structure but different residues;

[0085] Filtered output: Outputs a list of candidate sequences after modifying the CDR region.

[0086] Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essence of the present invention. Unless otherwise specified, the embodiments and features described in this application can be arbitrarily combined with each other.

Claims

1. A full-process antibody drug design and optimization system based on intelligent agents, characterized in that, The system includes: The intelligent interactive agent module is used to receive antibody design requirements input by users in the form of natural language, sequence files or structure files, and to identify user intentions, allocate tasks and extract parameters through an intelligent agent, and transform the user's unstructured design goals into executable antibody design constraints. The task planning and status management module is used to automatically construct the antibody design workflow according to the antibody design constraints, manage the execution status of each agent, task queue, user confirmation information, intermediate results and tool call results, and dynamically adjust the task plan according to the results of structure prediction, sequence generation, humanization evaluation or affinity prediction. The structural modeling and analysis module is used to call the protein structure prediction model to perform three-dimensional structural modeling on the input antigen sequence, antibody sequence, or antigen-antibody complex. It further analyzes the antigen epitope, antibody CDR region, antigen-antibody binding interface, spatial proximity, structural confidence, steric hindrance conflict, and binding hotspots, providing structural constraints for subsequent antibody backbone generation and sequence optimization. The de novo sequence generation module includes a backbone generation model and a sequence filling model. The backbone generation model generates a candidate antibody backbone structure that is spatially complementary to the antigen surface based on the antigen's three-dimensional structure, target epitopes or binding hotspots, and a preset CDR design region. The sequence filling model performs amino acid sequence filling, redesign, and optimization on the CDR region or a specified design region based on the candidate antibody backbone structure, antigen-antibody interface geometry, and local amino acid environment to generate multiple candidate antibody sequences. The structure validation and candidate screening module is used to re-predict the structure of antigen-antibody complexes from the generated candidate antibody sequences, and to screen and rank candidate antibodies based on CDR conformational retention, epitope contact, interface shape complementarity, structural confidence, spatial conflict, interface interaction and candidate sequence diversity. The sequence optimization and evaluation module includes a humanization model and a feasibility evaluation model. The humanization model performs homology searches, numbering, alignment, and frame region analysis on candidate antibody sequences against a human antibody germline gene database. It identifies the V-gene and J-gene with the highest similarity or best overall score as humanization reference templates and designs humanization mutations in frame regions or low-risk sites while preserving the conformation of the CDR region and key binding sites. The feasibility evaluation model assesses the immunogenicity risk, hydrophobicity, aggregation tendency, charge distribution, stability, expression feasibility, and sequence abnormality characteristics of candidate antibody sequences. The affinity prediction and mutation impact assessment module is used to call the protein-protein binding affinity prediction model to predict the binding free energy or affinity score of antibody-antigen complexes before and after humanization, compare the impact of humanization mutations, CDR redesign mutations or frame region mutations on binding ability, and combine structural stability, interfacial interaction and exploitability indicators to comprehensively rank the final candidate antibody sequences.

2. The agent-based end-to-end antibody drug design and optimization system according to claim 1, characterized in that, The antibody design constraints include one or more of the following: target antigen, antibody type, target epitope or binding hotspot, CDR region to be optimized, CDR length range, affinity optimization target, humanization target, and exploitability requirements.

3. The agent-based end-to-end antibody drug design and optimization system according to claim 1, characterized in that, The system also includes a report generation module, which automatically summarizes user input information, task plans, structural modeling results, epitope and hotspot analysis results, candidate antibody backbones, candidate antibody sequences, humanization schemes, affinity prediction results, exploitability assessment results, and recommended experimental verification schemes to generate a standardized antibody design optimization report.

4. A full-process antibody drug design and optimization method based on intelligent agents, characterized in that, The method includes the following steps: The intelligent agent receives the antibody design requirements input by the user, performs intent recognition, task classification and parameter extraction on the input, and automatically constructs an antibody design task plan. Structural modeling combined with hotspot analysis; Generate entirely new antibody backbones or optimize existing antibody backbones; CDR sequence filling and candidate sequence generation; Structural validation and affinity assessment; Human-centered design and development potential assessment; Candidate antibody ranking and result output; Generate a comprehensive design report.

5. The agent-based end-to-end antibody drug design and optimization method according to claim 4, characterized in that, The structural modeling and hotspot analysis steps specifically include: performing three-dimensional structural prediction on the input antigen sequence, antibody sequence, or antigen-antibody complex to obtain the antigen structure, antibody structure, or antigen-antibody complex structure; and analyzing the antigen epitopes, binding hotspots, CDR regions, antigen-antibody interfaces, spatial distances, steric hindrance conflicts, and structural confidence based on the structure to determine the structural constraints for subsequent antibody design.

6. The agent-based end-to-end antibody drug design and optimization method according to claim 4, characterized in that, The steps of generating a novel antibody backbone or optimizing an existing antibody backbone specifically include: generating a novel antibody backbone structure based on the target antigen structure, binding epitopes or hotspot residues, and a preset CDR design region, or redesigning one or more CDR regions in an existing antibody backbone to obtain a set of candidate antibody backbones that are spatially complementary to the target antigen.

7. The agent-based end-to-end antibody drug design and optimization method according to claim 4, characterized in that, The steps of CDR sequence filling and candidate sequence generation specifically include: based on the candidate antibody backbone structure, antigen-antibody interface geometric features and local structural environment, amino acid sequence filling or sequence optimization is performed on the specified CDR region to generate multiple candidate antibody sequences, and preliminary screening is carried out based on sequence rationality, structural compatibility, interface contact features and diversity.

8. The agent-based end-to-end antibody drug design and optimization method according to claim 4, characterized in that, The structure verification and affinity assessment steps specifically include: predicting the complex structure of the candidate antibody sequence and the target antigen; assessing whether the candidate sequence maintains the expected CDR conformation, whether it forms effective contact with the target epitope, and whether there is significant spatial conflict; further predicting the binding free energy or affinity score of the candidate antibody and the antigen, and assessing the impact of different candidate sequences and mutation sites on the binding ability.

9. The agent-based end-to-end antibody drug design and optimization method according to claim 4, characterized in that, The steps of humanization design and developmentability assessment specifically include: performing homology search and comparison between candidate antibody sequences and human antibody germline gene databases to determine humanization reference templates; designing humanization mutations for candidate antibodies while preserving CDR regions, key binding residues, and antigen-binding conformations, and assessing changes in affinity, immunogenicity risk, stability, hydrophobicity, aggregation tendency, expression feasibility, and overall developmentability of the sequences before and after humanization.

10. The agent-based end-to-end antibody drug design and optimization method according to claim 4, characterized in that, The steps of ranking and outputting the results of the candidate antibodies specifically include: ranking the candidate antibodies comprehensively based on structural quality, affinity prediction results, humanization degree, exploitability score, sequence diversity and user-defined optimization goals to form a recommended candidate antibody set.

Images