Computer-aided antibody off-target effect prediction method and system

Abstract

The invention discloses a computer-aided antibody off-target effect prediction method and system. The method mainly comprises the following steps: constructing an antibody-antigen structure and function annotation database, carrying out holoproteome structure search on an antibody by adopting a 3Di / aa space structure descriptor, and carrying out multi-dimensional comprehensive evaluation on a search result in combination with structural similarity. According to the method, high-flux and low-cost prediction of the antibody off-target effect can be realized, the limitation that the traditional method only depends on sequence alignment or single structure scoring is broken through, the prediction accuracy is improved, the risk caused by off-target is reduced, and the research and development period of antibody drugs is remarkably shortened.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer-aided drug research and development, in particular to an antibody off-target effect prediction method and system based on computer-aided, which is an antibody off-target effect prediction method and system based on spatial structure alignment and multi-parameter risk assessment. BACKGROUND

[0002] Antibody drugs have become one of the most important therapeutic means in the field of biological medicine due to their high specificity, high affinity and good targeting. Antibody drugs play an important role in the treatment of major diseases such as tumors, autoimmune diseases and infectious diseases. Compared with traditional small molecule drugs, antibody drugs have the advantages of high target specificity and relatively low side effects. However, the development of antibody drugs also has the characteristics of high investment, high risk and long cycle.

[0003] In the development of antibody drugs, off-target effects are an important factor affecting their safety and effectiveness. Off-target effects refer to the incorrect binding of non-target proteins or tissues by antibodies in addition to recognizing target antigens, which can cause serious adverse reactions. Therefore, how to discover and predict the off-target effects of antibodies in the early stage has become a key scientific problem that needs to be solved in the process of antibody development.

[0004] Currently, antibody off-target prediction mainly relies on in vitro experiments (such as tissue chip detection, protein chip screening, and immunoprecipitation) and in vivo animal model verification. Although these methods are reliable, they generally have the disadvantages of high cost, long cycle and low throughput, which cannot meet the needs of modern antibody drug rapid development.

[0005] With the rapid development of computational biology, structural biology and artificial intelligence technology, computational antibody off-target prediction methods have begun to attract widespread attention. Existing computational prediction methods mostly rely on sequence homology alignment or simple structure docking scoring, mainly through sequence alignment of target antigens and whole protein databases to screen potential off-target proteins with sequence similarity. However, this method can only capture linear sequence similarity and cannot effectively identify epitope cross-reactivity caused by spatial structure similarity. In addition, simple molecular docking scoring has the problems of insufficient accuracy and lack of global structure perception, which can easily cause high false positives or false negatives.

[0006] In recent years, deep learning techniques based on artificial neural networks have made significant breakthroughs in natural language processing, image recognition, and molecular structure prediction. For example, AlphaFold2 successfully achieved high-precision protein structure prediction, greatly promoting the development of structural biology. Deep learning techniques also show great potential in the field of biological medicine, and have better performance than traditional machine learning models in predicting drug-target interactions, protein interactions, drug discovery and virtual screening. Related studies have shown that deep learning methods based on structure can effectively capture the complex spatial relationships between proteins and improve the accuracy of interaction prediction.

[0007] With the continuous improvement of large-scale protein structure databases (such as the AlphaFold Protein Structure Database) and antibody databases (such as SAbDab), combined with deep learning-based structure alignment and multi-dimensional epitope similarity modeling, a new technical path for high-throughput and low-cost prediction of antibody off-target effects is provided. This method not only can identify off-targets caused by sequence homology, but also can capture complex off-target risks caused by epitope spatial structure similarity, charge distribution similarity, and surface chemical properties.

[0008] Therefore, there is an urgent need to develop an antibody off-target effect prediction method based on large-scale structure databases, combined with spatial structure descriptors (such as 3Di / aa, a structure encoding method that combines spatial topology and amino acid types, which can mine the spatial similarity between protein structures while ignoring sequence homology), and fusion of multi-dimensional physical and chemical parameters SUMMARY

[0009] To overcome the shortcomings of the prior art, the purpose of the present application is to provide a computer-aided antibody off-target effect prediction method, to improve the accuracy of antibody off-target prediction, to reduce the safety risks caused by off-targets in the research and development process, and to significantly shorten the research and development cycle of antibody drugs, and to reduce the research and development cost.

[0010] To solve the above problems, the technical solution adopted by the present application is as follows:

[0011] The embodiment of the present application provides a computer-aided antibody off-target effect prediction method, comprising,

[0012] Constructing an antibody-antigen structure and function annotation database: integrating data in antibody structure databases, protein structure databases and immunological epitope databases, constructing structure databases and function annotation databases, forming a database containing structure information and function annotation information;

[0013] The 3Di / aa spatial structure descriptor is used for the whole proteome search of the antibody: based on the structure database constructed above, the 3Di / aa descriptor is used for fast matching, and the protein surface fragments highly similar in spatial structure to the input antibody are screened out, and the spatial positions of the matched structure fragments and the corresponding functional information are output as the candidate list of potential off-target proteins;

[0014] The search results are comprehensively evaluated by multi-dimensional parameters: all potential off-target proteins are systematically analyzed by a double-index risk evaluation model, and the final off-target risk is calculated by a weighted linear model.

[0015] As a further preferred scheme, the structure database described in the embodiments of the present application is divided into four sub-databases;

[0016] The first sub-database contains the complete antibody variable region structure (VH / VL);

[0017] The second sub-database is a standardized CDR structure database formed according to the IMGT numbering rule, which extracts the complementarity determining region for each antibody sequence, including the heavy chain and the light chain;

[0018] The third sub-database focuses on the CDR region of the antibody heavy chain, and separately saves the structure information of CDRH1, CDRH2 and CDRH3;

[0019] The fourth sub-database extracts and saves the three-dimensional structure of the CDRH3 hypervariable region.

[0020] As a further preferred scheme, the structure database described in the embodiments of the present application includes the antibody variable region (VH / VL) structure, the spatial three-dimensional structure of the antigen epitope, and the three-dimensional coordinate information of the antibody-antigen complex; the structure database is derived from the crystal structure in the protein structure database or obtained by the three-dimensional structure prediction method based on the sequence.

[0021] As a further preferred scheme, the functional annotation database described in the embodiments of the present application contains antibody number, antibody name, antibody source and antigen number, antigen name, antigen function description and antigen germline source.

[0022] As a further preferred scheme, the database described in the embodiments of the present application synchronously stores the sequence and spatial three-dimensional coordinates of the antigen-antibody complex, the sequence and spatial three-dimensional coordinates of the antibody complementarity determining region (CDR), the sequence and spatial three-dimensional coordinates of the antibody heavy chain complementarity determining region (CDR H), and the sequence and spatial three-dimensional coordinates of the antibody heavy chain complementarity determining region 3 (CDR H3) structure description and chemical attribute description.

[0023] As a further preferred scheme, in the antibody off-target effect prediction method described in the embodiments of the present application, the full proteome search of the antibody using 3Di / aa spatial structure descriptors includes the following steps:

[0024] The input antibody or antigen-antibody complex structure is subjected to structural decomposition, and the key structural regions are extracted respectively;

[0025] The spatial structure descriptors are used to encode the structural regions, describe the three-dimensional spatial topology, amino acid sequence information, and the physical and chemical characteristics of the molecular surface;

[0026] The antibody complementarity determining region (CDR), heavy chain CDRH3 region, and the whole antibody are used as query templates to perform large-scale structural similarity search on the above-mentioned database, and a preliminary list of potential off-target proteins is obtained.

[0027] As a further preferred scheme, in the antibody off-target effect prediction method described in the embodiments of the present application, the search results are comprehensively evaluated by multi-dimensional parameters, including the following steps:

[0028] The TM-score is used to measure the spatial structure similarity between the antibody complementarity determining region (CDR) or antigen epitope and the potential matching region on the surface of the target protein, and a structural similarity score is calculated;

[0029] The spatial binding fitness score is calculated by molecular docking simulation using DockQ score or functionally equivalent methods;

[0030] Based on the structural similarity score and the spatial fitness score, a weighted linear model is used to calculate a comprehensive off-target risk score R.

[0031] As a further preferred scheme, in the antibody off-target effect prediction method described in the embodiments of the present application, the calculation formula of the comprehensive off-target risk score R calculated by the weighted linear model is:

[0032] R = w1·TM_Score + w2DockQ, wherein w1 is the weight of the structural similarity, w2 is the weight of the spatial fitness, and w1 + w2 = 1;

[0033] According to the numerical value of the comprehensive score R, the potential off-target proteins are divided into three risk levels:

[0034] When R > 0.8, it is determined as high risk, indicating that the structure is highly similar and the spatial fitness is good, and there is a significant off-target risk;

[0035] When 0.5 < R ≤ 0.8, it is determined as medium risk, with moderate structural similarity or partial spatial fitness, and there is a certain off-target probability;

[0036] When R≤0.5, it is determined as low risk, the structural topology is not similar, the spatial adaptability is poor, and the off-target probability is low.

[0037] As a further preferred scheme, in the antibody off-target effect prediction method described in the embodiments of the application, the weight is set as R=0.5(Query_TM_Score+Target_TM_Score) / 2+0.5DockQ or R=0.25Query_TM_Score+0.25Target_TM_Score+0.5DockQ.

[0038] The embodiments of the application also provide a computer-aided antibody off-target effect prediction system, comprising:

[0039] The antibody-antigen structure and function annotation database module comprises structure information and function annotation information of antibodies and antigens, and is used for providing a data basis for off-target effect prediction;

[0040] The antibody proteome search module is used for carrying out large-scale structural similarity search on the antibody binding region at the proteome level based on the 3Di / aa spatial structure descriptor, so as to identify potential off-target proteins;

[0041] The comprehensive evaluation module combines the TM-score to measure the spatial topological similarity of the antibody binding region and the potential epitope on the protein surface and the DockQ score to quantify the matching degree of the spatial binding interface, carries out off-target risk evaluation, and calculates the comprehensive off-target risk.

[0042] Compared with the prior art, the application has the beneficial effects that:

[0043] 1. The computer-aided antibody off-target effect prediction method described in the application breaks through the limitation of traditional methods that only rely on sequence alignment or simple structure scoring by fusing multi-dimensional evaluation of structural similarity and spatial adaptability, can more accurately identify potential off-target risks, and simultaneously has the advantages of high throughput and low cost, thereby providing efficient assistance for antibody drug research and development.

[0044] 2. The computer-aided antibody off-target effect prediction method described in the application directly captures spatial features such as three-dimensional topological structure, charge distribution and hydrophobicity of the protein surface based on the 3Di / aa spatial structure descriptor, rather than only relying on linear sequences, can identify off-target risks caused by similar spatial conformations, adopts a double-index evaluation model, combines a weighted linear model to calculate a comprehensive risk score, avoids the limitation of a single index, and improves the prediction reliability.

[0045] 3. Further, compared with protein chip screening, the computer-aided antibody off-target effect prediction method described in the present application can quickly process massive protein data and greatly improve the screening efficiency through large-scale structure search on the whole proteome database based on the computer method, meeting the needs of modern antibody drug rapid development; and without relying on expensive experimental equipment, reagents or animal models, the prediction can be completed only through computing resources, significantly reducing the research and development cost, and being particularly suitable for rapid risk screening in the early stage of research and development.

[0046] 4. The computer-aided antibody off-target effect prediction method described in the present application takes the spatial matching degree of antibody structure and target protein surface as the core evaluation index by fusing spatial topological similarity and binding interface adaptability. TM-score is used to measure the spatial topological similarity of antibody binding region and potential epitope on the protein surface, and DockQ score is used to quantify the matching degree of spatial binding interface. By combining these two core structural biology indexes, a comprehensive off-target risk score is generated through a simple and efficient weighted linear model. The scoring result can accurately reflect the structural similarity and potential binding possibility of the target protein and the antibody, thereby effectively predicting the off-target risk.

[0047] The application will be further described in detail below in combination with the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF DRAWINGS

[0048] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings needed in the embodiment 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 labor.

[0049] Figure 1 The flow chart of the computer-aided antibody off-target effect prediction method described in the embodiments of the present application. DETAILED DESCRIPTION

[0050] In order to make the purpose, technical scheme and advantages of the embodiments of the present application more clear, the technical scheme in the embodiments of the present application will be clearly and completely described below in combination with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0051] The term "comprising", as well as other similar terms such as "comprises" and "comprised of" used in the specification and claims, should not be interpreted as being restricted to the means listed thereafter; instead, such terms are to be interpreted as specifying the presence of stated features or steps but not to the exclusion in the absence of further recitations.

[0052] Embodiment 1

[0053] As Figure 1 shown, the embodiment of the present application provides a computer-aided antibody off-target effect prediction method. Taking 4zfg in the PDB database as an example, the antibody structure 5A12 Fab is predicted for potential target. 5A12 is a Fab fragment designed for bispecific antibody, which can recognize two different epitopes or two different targets at the same time. Bispecific antibody has important advantages in treatment, which can block two related pathogenic pathways at the same time. The binding antigen in 4zfg is ANGPT2, which is one of the members of the angiopoietin family, mainly involved in angiogenesis, vascular stability regulation and inflammatory response. The specific steps are as follows:

[0054] Step one: Construct antibody-antigen structure and function annotation database; including the construction of function annotation database and the systematic arrangement of structure database, aiming to provide high-quality structure basis and function support for the prediction of antibody off-target effect; specifically including,

[0055] S11, integrate the database, by integrating multiple public databases, including antibody structure database SAbDab, therapeutic antibody database Thera-SAbDab, antibody special database PLAbDab and immune epitope database IEDB, systematically collect antibody and antigen function annotation information and structure information; wherein the function annotation information contains the number, name and source of each antibody, and the number, name, function description, germline source, protein function category, tissue expression profile, homology family information of each antigen.

[0056] S12, obtain structure information, including three ways of obtaining and standardizing the structure by directly extracting high-quality antibody-antigen complex crystal structure from PDB database, using EsmFold to perform high-precision structure prediction based on protein sequence, and using BoltzFold to perform high-precision structure prediction based on protein sequence; the standardized structure corresponds to the antibody ID and antigen ID in the function annotation database one by one, ensuring the consistency and traceability of the data.

[0057] S13, database establishment, construct structure database and functional annotation database according to functional annotation information and structure information; further divide the structure database into four sub-databases according to different functional domains; first, Antibody_db, containing complete antibody variable region structure (VH / VL), used for similarity analysis at the whole structure level; second, CDR_imgt_db, according to IMGT numbering rules, extract the complementarity determining region (CDR1, CDR2 and CDR3) of each antibody sequence, including heavy chain and light chain, to form a standardized CDR structure database; third, CDRH_imgt_db, this database further focuses on the CDR region of antibody heavy chain, separately saves the structure information of CDRH1, CDRH2 and CDRH3, and provides more accurate search support for the binding mechanism dominated by heavy chain; fourth, CDRH3_imgt_db, specially extracts and saves the three-dimensional structure of CDRH3 hypervariable region; because CDRH3 has the most significant spatial conformation change in antibody-antigen binding, and is also the key region determining antibody specificity, it is separately built into a database, which can effectively identify those high-risk off-target proteins caused by accidental structural similarity of CDRH3. The functional information database can support the functional risk identification of off-target proteins, and provide tagged auxiliary information for subsequent spatial structure similarity search.

[0058] Step two: perform whole proteome search on antibodies using 3Di / aa spatial structure descriptor; based on the structure database constructed above, perform fast matching using 3Di / aa descriptor, screen out protein surface fragments highly similar in spatial structure to the input antibody, and output the spatial position of the matched structure fragments and the corresponding functional information as the candidate list of potential off-target proteins; including

[0059] S21, decompose the structure of the input antibody or antigen-antibody complex, and divide it into complete antibody structure, complementarity determining region (CDR) in antibody variable region (VH / VL), antibody heavy chain complementarity determining region (CDRH), and CDRH3 hypervariable region.

[0060] S22, take the CDR region, CDRH region and CDRH3 region of the antibody as the main query template, and perform whole proteome spatial structure similarity search on the four sub-databases (Antibody_db, CDR_imgt_db, CDRH_imgt_db and CDRH3_imgt_db) and the whole human protein structure database established in step one. The search process is based on high-performance structure alignment tool, and uses 3Di / aa descriptor for fast matching to screen out protein surface fragments highly similar in spatial structure to the input antibody.

[0061] S23, for each matched protein, record its unique ID, protein name, functional annotation, matching score and specific spatial matching position; the matching score uses TM-score as the measurement standard, respectively calculates Query TM-score (i.e. the fitting degree of antibody structure mapping to target protein) and Target TM-score (i.e. the fitting degree of target protein surface structure mapping to antibody structure); in addition, output the spatial position of the matched structure fragment and the corresponding functional information as the candidate list of potential off-target proteins; the specific search results of this step are listed in Table 1.

[0062] Table 1: Partial results of Antibody_db sub-database search

[0063]

[0064] Step three: comprehensive evaluation of search results by multi-dimensional parameters, all potential off-target proteins are systematically analyzed by a two-index risk evaluation model, and the final off-target risk is calculated by a weighted linear model; specifically including,

[0065] S31, based on structural topological similarity, calculate the structural similarity score; this score takes TM-score as the core index, comprehensively considers the two-way fitting degree of Query structure and Target structure, and is used to quantify the spatial structural similarity between antibody complementarity determining region (CDR) or key binding region (such as CDRH3) and potential off-target region. TM-score takes value in the range of 0 to 1, the higher the value, the stronger the structural similarity.

[0066] S32, calculate the spatial binding fitness score; DockQ score is used as the measurement index to evaluate the binding possibility of antibody CDR structure and target protein surface matching region in spatial conformation; DockQ score combines multiple spatial parameters such as interface contact score (Fnat), interface RMSD and ligand RMSD, which can accurately reflect the spatial fitness between the two; DockQ takes value in the range of 0 to 1, the higher the value, the better the spatial adaptability, and has higher potential binding risk.

[0067] S33, according to the two scores above, the final off-target risk score R is calculated by using a weighted linear model, the formula is: R = w1 TM_Score + w2 DockQ, wherein w1 is the weight of structural similarity, w2 is the weight of spatial fitness, and w1 + w2 = 1. According to experience and historical off-target verification data, it is recommended to use w1 = 0.5, w2 = 0.5, and the calculation formula is R = 0.5 (Query_TM_Score + Target_TM_Score) / 2 + 0.5 DockQ, or another weighted average formula R = 0.25 Query_TM_Score + 0.25 Target_TM_Score + 0.5 DockQ.

[0068] According to the three structural source paths in step one, three comprehensive risk scores R are obtained respectively, and the average value is taken as the final risk score R (some missing values in Experiential indicate that there is no diffraction crystal structure of the protein in the PDB database).

[0069] According to the results of the comprehensive score R, a risk classification system is established. Specifically, when R > 0.8, it is determined as high off-target risk, indicating that the structure is highly similar and the spatial fitness is good, and there is a significant off-target possibility; when 0.5 < R ≤ 0.8, it is determined as medium risk, indicating that the target protein and the antibody have certain structural fitting and spatial fitness, and have potential off-target risk; when R ≤ 0.5, it is determined as low risk, indicating that the structural similarity and spatial fitness are both low, and the off-target probability is small.

[0070] The comprehensive risk assessment results are listed in Table 2, which shows the Query TM-score, Target TM-score, DockQ score, final comprehensive risk score R and corresponding risk classification level of each potential off-target protein. Through Table 2, it can be directly observed which proteins belong to high risk, medium risk or low risk, effectively supporting the quantitative evaluation and risk avoidance of off-target effect in antibody design process.

[0071] Table 2: Partial high-risk results of comprehensive risk assessment

[0072]

[0073]

[0074] The comprehensive score R of antibody off-target risk prediction is an important standard for evaluating the accuracy of the screening method, and the value of R ranges from 0 to 1. The closer the value is to 1, the more accurate the prediction model is in judging the off-target risk. As shown in the experimental results in Tables 1 and 2, the present application has a higher accuracy in predicting antibody off-target effects, while maintaining the advantages of high throughput and fast calculation. The double-index scoring method based on structural similarity (TM-score) and spatial combination fitness (DockQ) used in the present application can effectively screen out proteins with potential off-target risks from a large proteome database. The higher the accuracy of the scoring method, the more accurately it can identify potential off-target targets that actually exist in spatial fitting and structural similarity, thereby effectively reducing the off-target risk and avoiding false positive or false negative errors. This not only helps to improve the success rate of antibody drug design, but also significantly reduces the safety risk in subsequent animal experiments and clinical verification.

[0075] The above embodiments are only preferred embodiments of the present application, and cannot be used to limit the scope of protection of the present application. Any non-essential changes and substitutions made by those skilled in the art on the basis of the present application are within the scope of the present application.

Claims

1. A computer-aided based method for predicting antibody off-target effects, characterized in that, comprising, constructing antibody-antigen structure and function annotation database: integrating data in antibody structure database, protein structure database and immunological epitope database, constructing structure database and function annotation database, forming a database containing structure information and function annotation information; performing whole proteome search on antibodies using 3Di / aa spatial structure descriptor: based on the structure database constructed above, using 3Di / aa descriptor for fast matching, screening out protein surface fragments highly similar in spatial structure to the input antibody, outputting the spatial position of the matched structure fragments and the corresponding function information as the candidate list of potential off-target proteins; comprehensively evaluating the search results through multi-dimensional parameters: all potential off-target proteins are analyzed by a two-index risk evaluation model, and the final off-target risk is calculated by a weighted linear model.

2. The antibody off-target effect prediction method of claim 1, wherein, The structure database is divided into four sub-databases. The first sub-database contains the complete antibody variable region structure (VH / VL); The second sub-database is a standardized CDR structure database formed according to IMGT numbering rules, which extracts the complementarity determining region for each antibody sequence, including heavy chain and light chain; The third sub-database focuses on the CDR region of antibody heavy chain, and separately saves the structure information of CDRH1, CDRH2 and CDRH3; The fourth sub-database extracts and saves the three-dimensional structure of CDRH3 hypervariable region.

3. The antibody off-target effect prediction method of claim 1, wherein, The structure database includes antibody variable region (VH / VL) structure, spatial three-dimensional structure of antigen epitope, and three-dimensional coordinate information of antibody-antigen complex. The structure database is derived from crystal structures in protein structure database or three-dimensional structures obtained by sequence-based structure prediction method.

4. The antibody off-target effect prediction method of claim 1, wherein, The function annotation database contains antibody number, antibody name, antibody source, antigen number, antigen name, antigen function description and antigen germline source.

5. The antibody off-target effect prediction method of claim 1, wherein, The database synchronously stores the sequence and its spatial three-dimensional coordinates of the antigen-antibody complex, the sequence and its spatial three-dimensional coordinates of the antibody complementarity determining region (CDR), the sequence and its spatial three-dimensional coordinates of the antibody heavy chain complementarity determining region (CDR H), and the sequence and its spatial three-dimensional coordinates of the antibody heavy chain complementarity determining region 3 (CDRH3).

6. The antibody off-target effect prediction method according to any one of claims 1 to 5, characterized in that, Performing whole proteome search on antibodies using 3Di / aa spatial structure descriptor includes the following steps: Split the input antibody or antigen-antibody complex structure, and extract the key structure region respectively; For the above structure region, use spatial structure descriptor to code it, describe three-dimensional spatial topology, amino acid sequence information, and physical and chemical characteristics of molecular surface; Taking antibody complementarity determining region (CDR), heavy chain CDRH3 region and antibody as a whole as a query template, large-scale structure similarity search is performed on the above database to obtain a preliminary list of potential off-target proteins.

7. The antibody off-target effect prediction method according to any one of claims 1 to 5, characterized in that, Comprehensively evaluating the search results through multi-dimensional parameters includes the following steps: TM-score is used to measure the structural similarity between the antibody CDR or antigen epitope and the potential matching region on the surface of the target protein; DockQ score or functionally equivalent methods are used to calculate the spatial binding fitness score through molecular docking simulation; Based on the structural similarity score and spatial fitness score, a weighted linear model is used to calculate the comprehensive off-target risk score R.

8. The antibody off-target effect prediction method of claim 7, wherein, The formula for calculating the comprehensive off-target risk score R by the weighted linear model is: R = w1·TM_Score + w2DockQ, where w1 is the weight of structural similarity, w2 is the weight of spatial fitness, and w1 + w2 = 1; According to the value of the comprehensive score R, the potential off-target proteins are divided into three risk levels: When R > 0.8, it is determined as high risk, indicating high structural similarity and good spatial adaptability, with significant off-target risk; When 0.5 < R ≤ 0.8, it is determined as medium risk, with moderate structural similarity or partial spatial adaptability, with certain off-target probability; When R ≤ 0.5, it is determined as low risk, with dissimilar structure topology and poor spatial adaptability, with low off-target probability.

9. The antibody off-target effect prediction method of claim 8, wherein, The weight setting is R = 0.5(Query_TM_Score + Target_TM_Score) / 2 + 0.5DockQ, or the weighted average form R = 0.25Query_TM_Score + 0.25Target_TM_Score + 0.5DockQ.

10. A computer-aided based antibody off-target effect prediction system, characterized in that, It includes: Antibody-antigen structure and function annotation database module, containing the structural information and functional annotation information of antibodies and antigens, providing data basis for off-target effect prediction; Antibody proteome search module, based on 3Di / aa spatial structure descriptor, large-scale structural similarity search is carried out on the antibody binding region at the whole proteome level to identify potential off-target proteins; Comprehensive evaluation module, combining TM-score to measure the spatial topological similarity between antibody binding region and potential epitope on the surface of protein and DockQ score to quantify the matching degree of spatial binding interface, to evaluate the off-target risk and calculate the comprehensive off-target risk.

Images