A protein binder intelligent design method oriented to target protein function site blocking and fusion off-target risk assessment

Abstract

The application discloses an intelligent design method for protein binders aiming at blocking target protein functional sites and evaluating off-target risks, and belongs to the technical field of artificial intelligence protein design. In order to solve the problems that in the existing protein binder design method, candidate binders are difficult to ensure effective blocking of the key functional regions of target proteins, and systematic off-target risk evaluation is lacking, resulting in insufficient function effectiveness and safety, the application comprises the following steps: obtaining target protein information, performing feature analysis, identifying the key functional regions, and constructing the functional site blocking constraints; generating candidate protein binders, and performing sequence optimization on the non-interface regions; predicting the complex structure, and calculating the function blocking score; constructing a non-target protein set, predicting the cross-binding ability of the candidate protein binders and the non-target proteins, and calculating the off-target risk score; determining the preferred candidate, and outputting the sequence information, the structure information and the score report. The application is used for designing candidate binders with high blocking ability and low off-target risk for the functional sites of target proteins.

Description

Technical Field

[0001] This invention relates to a method for intelligent design of protein conjugates that blocks target protein functional sites and assesses off-target risks, belonging to the field of artificial intelligence protein design technology. Background Technology

[0002] Protein conjugates are a class of functional molecules capable of selectively binding to specific target proteins. They can regulate related biological functions by occupying or blocking key regions of the target protein, thus affecting the binding between the target protein and its natural ligands, substrates, receptors, or interacting proteins. Antibodies, nanobodies, mini proteins, artificially designed binding proteins, and scaffold proteins are all important candidate forms of protein conjugates. Because protein conjugates typically have large binding interfaces and strong targeting capabilities, they have significant application value in receptor-ligand blocking, protein-protein interaction inhibition, enzyme activity regulation, and signaling pathway intervention.

[0003] Traditional protein conjugate design primarily relies on experimental screening methods such as immune acquisition, phage display, or yeast display, which are resource-intensive, have long production cycles, and are costly. In recent years, with the development of deep learning structure prediction models, protein language models, and generative protein design methods, the computational design efficiency of protein conjugates based on structural information and artificial intelligence has significantly improved. Existing technologies can generate candidate conjugates based on the three-dimensional structure of the target protein, predict the complex conformation, and optimize the stability, solubility, and structural adaptability of candidate sequences, thereby reducing the cost of traditional experimental screening.

[0004] However, the existing computational design process has the following shortcomings:

[0005] First, existing methods typically focus more on the binding ability between candidate conjugates and target proteins, such as binding energy, interfacial area, contact number, structural confidence score, or complex prediction score. While these indicators can reflect whether a candidate conjugate is likely to stably bind to the target protein, they cannot guarantee that the binding site has a clear functional blocking effect. Even if a candidate conjugate has a high binding score, it may bind to the non-functional surface of the target protein, far from the ligand binding site, protein-protein interaction interface, enzyme active site, or key signal transduction region, thus making it difficult to effectively block disease-related functions.

[0006] Second, existing methods are not precise enough in assessing functional site blockade. Although some methods involve calculating the probability of protein-ligand binding blockade, they usually use simplified models based on atomic coverage, which fail to fully consider the multi-level characteristics of functional sites (such as residue conservation, interfacial contact frequency, mutation effects, energy contribution, etc.) and make it difficult to construct differentiated blocking constraints for different key functional regions (such as receptor binding sites, enzyme active sites, signal transduction regions, etc.).

[0007] Third, the specificity and safety assessment of protein conjugates are insufficient. Many disease-related target proteins share similar surface structures or conserved functional regions with their homologs, family members, or structurally similar proteins. If candidate conjugates are optimized only for the target protein, they may cross-bind with other non-target proteins simultaneously. This is especially problematic when non-target proteins are highly expressed in normal tissues or participate in important physiological functions, potentially leading to off-target toxicity or interference with normal signaling pathways. While some existing methods involve off-target protein recognition or anti-off-target effect design, most are aimed at small molecule drugs rather than protein conjugates, and they fail to organically integrate off-target risk assessment with functional site blocking design.

[0008] Fourth, there is a lack of multi-target comprehensive screening mechanisms. The clinical application of protein conjugates requires simultaneously meeting multiple objectives, such as high functional blocking ability, high target specificity, good structural stability, good developability, and low off-target risk. Existing design processes often focus on a single or a few indicators, making it difficult to effectively balance and comprehensively prioritize multiple objectives.

[0009] Therefore, protein conjugate design cannot rely solely on binding strength or structural stability indicators. It is also necessary to determine whether candidate conjugates cover key functional regions of the target protein, whether they can effectively block ligand binding, protein-protein interaction interfaces, enzyme active sites, or signal transduction regions, and to further assess the potential risk of misbinding with homologous proteins, structurally similar proteins, or normal human proteins.

[0010] Therefore, it is necessary to establish a smart design method for protein conjugates that integrates functional site identification, functional blocking constraints, functional blocking scoring, off-target risk assessment, and multi-target screening to improve the functional effectiveness, target specificity, and drug development feasibility of candidate protein conjugates. Summary of the Invention

[0011] The purpose of this invention is to address the problems in existing protein conjugate design methods, such as the difficulty in ensuring effective blocking of key functional regions of target proteins by candidate conjugates and the lack of systematic off-target risk assessment, which leads to insufficient functional effectiveness and safety. This invention provides a smart protein conjugate design method that addresses the blocking of target protein functional sites and incorporates off-target risk assessment.

[0012] The present invention discloses a smart design method for protein conjugates that blocks target protein functional sites and assesses off-target risks, comprising the following steps:

[0013] Step 1: Obtain disease-related target protein information, perform feature analysis on the target protein information, identify key functional regions, and construct functional site blocking constraints based on the key functional regions;

[0014] Step 2: Generate candidate protein conjugates based on the functional site blocking constraints. Divide interface residues and non-interface residues according to their spatial contact relationship with the target protein. While keeping the key residues involved in the functional site contact in the interface residues unchanged, perform sequence optimization on the non-interface region.

[0015] Step 3: For the candidate protein conjugates obtained in Step 2, predict the structure of the complexes they form with the target protein, and calculate the functional blocking score based on the complex structure; the functional blocking score is used to quantify the blocking ability of the candidate protein conjugates on the key functional regions of the target protein.

[0016] Step 4: Construct a set of non-target proteins, predict the cross-binding ability of the candidate protein conjugates obtained in Step 2 with the non-target proteins, and calculate the off-target risk score by combining the sequence homology and structural similarity between the non-target proteins and the target proteins, as well as the expression risk of the non-target proteins in normal tissues.

[0017] Step 5: Based on the functional blocking score obtained in Step 3 and the off-target risk score obtained in Step 4, perform multi-target comprehensive screening by combining the target binding score, structural stability score and exploitability score of the candidate protein conjugate, determine the preferred candidate, and output the sequence information, structural information and various score reports of the preferred candidate protein conjugate.

[0018] Preferably, the disease-related target protein information in step 1 includes one or more of the following: target protein amino acid sequence, three-dimensional structure, PDB file, predicted structure, protein name, species origin, disease type, functional category, natural ligand, substrate, receptor, interacting protein, and known mutation site information.

[0019] The feature analysis includes analyzing the spatial coordinates, secondary structure, surface accessibility, charge, hydrophobicity, and conservation characteristics of target protein residues to form a structured data object of the target protein.

[0020] Preferably, the functional key region described in step 1 includes one or more of the following: receptor binding site, ligand binding site, protein-protein interaction interface, enzyme active site, substrate binding pocket, signal transduction key region, allosteric regulation region, and user-specified hotspot residues.

[0021] The key functional regions are determined through one or more of the following methods: user specification, structural annotation parsing, extraction of known complex interfaces, pocket identification, conservation analysis, or hotspot residue prediction.

[0022] Functional site set Represented as:

[0023] ;

[0024] Hotspot residue set Represented as:

[0025] ;

[0026] in, It represents the set of all residues of the target protein. Indicates the first One target protein residue, Indicates functional locus score, Indicates hotspot residue score, and These represent the functional site threshold and the hotspot residue threshold, respectively.

[0027] Preferably, the hotspot residue score is calculated based on at least two of residue conservation, interfacial contact frequency, mutation effect, and energy contribution, and is expressed as follows:

[0028] ;

[0029] in, Indicates the first Conservation score of each residue, Indicates the frequency of interface contact. This indicates that the mutation affects the score. Indicates the energy contribution score. , , , These are the weight parameters.

[0030] Preferably, the specific method for constructing functional site blocking constraints based on the functionally critical regions in step 1 includes:

[0031] Functional site blocking constraints are generated based on the spatial coordinates, surface accessibility, charge distribution, hydrophobicity, secondary structure environment, and spatial extent of residues in functionally critical regions.

[0032] Functional site blocking constraints include one or more of the following: target contact residue set, preferred coverage region, exclusion region, binding direction constraint, spatial distance constraint, spatial occlusion constraint, interface shape complementarity constraint, and candidate binding posture restriction;

[0033] Among them, priority coverage areas Represented as:

[0034] ;

[0035] in, Represents any point in three-dimensional space. Indicates hotspot residues spatial coordinates, Represents Euclidean distance. Indicates the blocking radius.

[0036] Preferably, the specific method for generating candidate protein conjugates based on the functional site blocking constraint in step 2 includes:

[0037] The coverage constraints of candidate protein conjugates on functional sites are expressed as follows:

[0038] ;

[0039] in, Indicates candidate protein conjugates. Indicates the first in the candidate protein conjugate One residue, Indicates the contact distance threshold;

[0040] Candidate protein conjugates are generated using structure prediction models, diffusion generation models, sequence design models, protein language models, and structure generation models. The generated candidate protein conjugates satisfy the coverage constraints.

[0041] The generation of the candidate protein conjugate includes backbone generation, sequence generation, and binding posture generation;

[0042] The set of candidate protein conjugates is represented as follows:

[0043] ;

[0044] in, Indicates the first One candidate protein conjugate, Indicates the number of candidates;

[0045] The candidate generation loss function is expressed as:

[0046] ;

[0047] in, This indicates the distance penalty for candidate protein binders deviating from the functional site region. This indicates a penalty for insufficient coverage of functional sites. This indicates a penalty for insufficient contact with hotspot residues. Indicates spatial conflict penalty, This indicates a penalty for insufficient structural stability. to These are weight parameters;

[0048] The candidate protein conjugates include antibody fragments, nanobodies, mini proteins, artificially designed binding proteins, or scaffold proteins.

[0049] Preferably, the specific method for dividing interface residues and non-interface residues based on their spatial contact relationship with the target protein in step 2 includes:

[0050] The set of interfacial residues of candidate protein-conjugates is represented as:

[0051] ;

[0052] The set of non-interface residues is represented as:

[0053] ;

[0054] in, Indicates target protein, This represents the set of all residues in a candidate protein-binding compound. This represents the set of interface residues of candidate protein-binding compounds. This represents the set of non-interfacial residues of candidate protein-conjugates. Indicates the interface distance threshold;

[0055] Fixed constraints are set for key interface residues involved in the contact of functionally critical regions:

[0056] ;

[0057] in, Represents the set of key residues at the interface. Indicates the residue type before optimization. Indicates the optimized residue type;

[0058] For sequence redesign of non-interface residues, the optimization objective is expressed as:

[0059] ;

[0060] in, Indicates the stability score. Indicates solubility score. Indicates the feasibility score. This indicates a risk of gathering. Indicates potential development risk. to These are the weight parameters.

[0061] Preferably, step 3, which involves predicting the complex structure of each candidate protein conjugate with the target protein and calculating a functional blocking score based on the complex structure, specifically includes:

[0062] The complex structure prediction output includes one or more of the following: binding attitude, interface residues, residue contact map, number of interface contacts, complex structure confidence, interface confidence, geometric complementarity, interface area, attitude consistency, and degree of spatial conflict.

[0063] The functional blocking score is calculated based on at least two of the following: functional site coverage, hotspot residue contact score, spatial occlusion score, interfacial area score, and binding posture stability score, and is expressed as follows:

[0064] ;

[0065] in, Indicates the first Functional blocking scores for candidate protein conjugates Indicates the weighted functional site coverage. Indicates the contact score of hotspot residues. Indicates spatial occlusion score. The score represents the interface area. This indicates that the attitude stability score is combined. to These are the weight parameters.

[0066] Preferably, the set of non-target proteins in step 4 includes one or more of the following: proteins that have sequence homology with the target protein, proteins that have structural similarity with the target protein, proteins that have similar surface pockets or domains, and proteins expressed in normal human tissues.

[0067] Step 4 involves considering the sequence homology and structural similarity between the non-target protein and the target protein, as well as the expression risk of the non-target protein in normal tissues, to calculate an off-target risk score, expressed as follows:

[0068] ;

[0069] in, Indicating candidate protein conjugates Regarding the first Non-target protein Off-target risk score, This indicates the predicted binding ability of candidate protein conjugates to non-target proteins. This indicates the sequence similarity between non-target proteins and target proteins. Indicating structural similarity, This indicates that a normal organization is expressing risk. to These are weight parameters;

[0070] For multiple non-target proteins, the overall off-target risk score is obtained by maximizing the value or by weighted summation:

[0071] ;

[0072] or: ;

[0073] in, Indicates the first Risk weights for non-target proteins.

[0074] Preferably, the method for determining preferred candidates by performing multi-target comprehensive screening based on functional blocking score and off-target risk score, combined with the target binding score, structural stability score, and exploitability score of candidate protein conjugates, as described in step 5 is as follows:

[0075] The overall score is expressed as follows:

[0076] ;

[0077] in, Indicates the first A comprehensive score for each candidate protein conjugate Indicates the functional blocking score. Indicates the target binding score, Indicates the structural stability score. Indicates the potential for development score. Indicates the off-target risk score. to These are weight parameters;

[0078] Candidate protein conjugates are sorted, graded, or eliminated based on the comprehensive score and a preset threshold, wherein candidate protein conjugates with off-target risk scores exceeding the preset threshold are downgraded or eliminated.

[0079] Step 5 outputs the sequence information, structural information, and various scoring reports of the preferred candidate protein conjugates, which specifically include:

[0080] The amino acid sequence, three-dimensional structure, complex structure, binding interface residues, functional site coverage, hotspot residue contact, spatial occlusion, various scores, candidate classification, experimental validation priority, and interpretation report of the candidate protein conjugate.

[0081] Advantages of this invention: By introducing functional site blocking constraints, this invention enables candidate protein conjugates to preferentially cover or contact key functional regions of target proteins, reducing candidates that "can bind but cannot block function"; through a functional blocking scoring system, functional site coverage, hotspot residue contact, spatial occlusion, interface area, and posture stability are quantitatively evaluated; through an off-target risk assessment mechanism, the potential misbinding risk of homologous proteins, structurally similar proteins, and normally expressed human proteins is incorporated into the screening process; through multi-target comprehensive screening, candidate protein conjugates with high functional blocking ability, high target specificity, good structural stability, good developability, and low off-target risk can be obtained. Attached Figure Description

[0082] Figure 1 This is a flowchart of a method for intelligent design of protein conjugates that blocks target protein functional sites and assesses off-target risks, as described in this invention.

[0083] Figure 2 This is a schematic diagram of the system module architecture for implementing a smart design method of protein conjugates that targets the functional sites of target proteins and integrates off-target risk assessment.

[0084] Figure 3 This is a schematic diagram of functional site blocking constraints;

[0085] Figure 4 This is a diagram illustrating fixed interface and non-interface optimization.

[0086] Figure 5 This is a schematic diagram of the complex contact and blocking score. Detailed Implementation

[0087] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0088] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other.

[0089] The present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the scope of the invention.

[0090] Example 1:

[0091] The following is combined with Figures 1-5 This embodiment describes a smart design method for protein conjugates that blocks target protein functional sites and assesses off-target risks. Figure 1 This is a flowchart of a protein-conjugate intelligent design method, such as... Figure 2 The diagram shown illustrates the system module architecture for implementing a smart design method for protein conjugates, including the following steps:

[0092] Step 1: Obtain disease-related target protein information, perform feature analysis on the target protein information, identify key functional regions, and construct functional site blocking constraints based on the key functional regions.

[0093] Take human epidermal growth factor receptor (EGFR), a target associated with non-small cell lung cancer, as an example. EGFR has the UniProt number P00533, a total length of 1210 amino acid residues, and is associated with diseases such as non-small cell lung cancer.

[0094] The system obtains one or more of the following information about the target protein: amino acid sequence, three-dimensional structure (PDB file), protein name, species origin, disease type, functional category, natural ligand, substrate, receptor, interacting protein, and known mutation sites.

[0095] In this embodiment, the system obtained the amino acid sequence of EGFR, the PDB structure file, and information on known drug resistance mutations.

[0096] The system standardizes the input data and analyzes the spatial coordinates, secondary structure, surface accessibility, charge, hydrophobicity, and conservation characteristics of target protein residues to form a structured data object of the target protein.

[0097] All residues of EGFR are represented as: ;

[0098] The feature vector of each residue is represented as: ;

[0099] in, The score indicates the structure or function annotation. Indicates the pocket recognition score. Indicates the frequency of interface contact. This indicates that the mutation affects the score. The score indicates the level of conservatism. All scores are normalized to between 0 and 1, with higher values ​​indicating that the residue is more likely to belong to a functionally critical region.

[0100] The system identifies functionally critical regions on the EGFR.

[0101] The key functional regions include one or more of the following: receptor binding site, ligand binding site, protein-protein interaction interface, enzyme active site, substrate binding pocket, signal transduction key region, allosteric regulation region, and user-specified hotspot residues.

[0102] Functionally critical regions are determined through one or more of the following methods: user specification, structural annotation parsing, extraction of known complex interfaces, pocket identification, conservation analysis, or hotspot residue prediction.

[0103] When performing structural annotation analysis, if a residue is annotated as an ATP-binding core residue by the PDB structure, its structural annotation score is recorded as follows: If it is not a core residue but is near the ATP-binding pocket, it can be denoted as... .

[0104] In this embodiment, K721 is labeled as an ATP-binding core residue, therefore .

[0105] When performing known complex interface extraction, the system calculates the target protein residues. With natural ligands or interacting protein residues Minimum distance between: .

[0106] If satisfied (Settings in this embodiment) If so, then the target protein residues are considered to be... It forms contact with natural ligands or interacting proteins.

[0107] Interface contact frequency is expressed as: ;

[0108] in, This is an indicator function that takes the value 1 if the condition is met, and 0 otherwise. This indicates the number of ligand or interacting protein residues.

[0109] The settings interface is determined as follows If a certain EGFR residue If it forms a contact within 5 Å with 8 of the 10 residues in the natural ligand, then: ;

[0110] Calculate pocket score: ;

[0111] set up As a threshold, the pocket score satisfies The residues were identified as pocket-related residues.

[0112] For example, in the EGFR embodiment, the system obtains pocket scores for several residues as shown in Table 1:

[0113] Table 1

[0114]

[0115] In this embodiment, K721, T790, and V742 were identified as pocket-related residues with pocket scores of 0.86, 0.84, and 0.75, respectively.

[0116] During conservation analysis, the system performs multiple sequence alignment on the target protein sequence and calculates the amino acid distribution entropy at each residue position: .

[0117] in, Indicates the first The first position The frequency of occurrence of certain amino acids.

[0118] Then convert it into a conservative score: .

[0119] The closer it is to 1, the more conservative the position. The closer it is to 0, the less conservative the position is.

[0120] Set the threshold to Then conservative score The residues were identified as conserved functional residues.

[0121] Substituting specific data into the example, the system scored several residues within the EGFR kinase domain, and the results are shown in Table 2:

[0122] Table 2

[0123]

[0124] The system settings function site scoring formula is as follows: ;

[0125] The system will meet The residues were identified as functional site residues.

[0126] Meanwhile, the system further calculates hotspot residue scores: ;

[0127] in, The score represents the energy contribution of the residue.

[0128] The system will meet The residues were identified as hotspot residues.

[0129] For example, T790: ;

[0130] but: ;

[0131] Therefore, T790 was identified as a hotspot residue.

[0132] Ultimately, the system obtains the set of functional sites for EGFR: ;

[0133] Hotspot residue set: ;

[0134] For residues whose scores do not reach the threshold, such as V742, they can be recorded as auxiliary functional regions: .

[0135] Constructing functional site blocking constraints:

[0136] The system establishes a spatial model of functional sites based on the three-dimensional coordinates of each functional residue. The Cα atom coordinates of each functional residue are shown in Table 3.

[0137] Table 3

[0138] The system calculates the center of the hotspot region based on the coordinates of the core hotspot residues: ;

[0139] Substitute the coordinates of the five hotspot residues above:

[0140] ;

[0141] This point serves as the spatial center of the EGFR functional blocking region, defining the location for the formation of candidate protein binders.

[0142] The system sets hotspot residue sets as the target residue sets that candidate protein binders must preferentially contact:

[0143] ;

[0144] At the same time, an importance weight is assigned to each hotspot residue:

[0145] ;

[0146] The system prioritizes the space within a certain radius around hotspot residues as the blocking region. ;

[0147] in, Represents any point in three-dimensional space. Represents Euclidean distance. Indicates the blocking radius.

[0148] Combination Figure 3 ,set up For auxiliary residue V742, set the auxiliary coverage radius. This allows candidate protein conjugates to extend appropriately into the region near the ATP-binding pocket while covering core hotspot residues. For candidate protein conjugates... any residue in The system calculates its relationship with EGFR functional residues. Minimum distance:

[0149] ;

[0150] When satisfied At this time, the candidate protein binder forms an effective proximity relationship with the functional residue. In this embodiment, a contact distance threshold is set. ,like , , , , Then the candidate protein conjugate A total of 4 core hotspot residues were contacted.

[0151] The system defines whether each functional residue is overridden:

[0152] ;

[0153] Further calculate the weighted coverage ratio: ;

[0154] For candidate protein conjugates The coverage result is as follows:

[0155] ;

[0156] Substitute the weights: ;

[0157] System settings minimum coverage threshold ,because Therefore, candidate protein conjugates The system satisfies functional site coverage constraints. In addition to coverage, the system also sets constraints on the number of hotspot residue contacts:

[0158] ;

[0159] And require In the example above ,therefore, The constraint on the number of hotspot residues is met.

[0160] Candidate protein conjugates The center of the interface is: ; Indicates the number of interface residues.

[0161] The system requires that the distance between the center of the candidate bonding interface and the center of the hot spot region meets the following requirements. For example, candidate protein conjugates The center of the interface is

[0162] but: ;

[0163] because ,therefore Satisfy the combined attitude center distance constraint.

[0164] Candidate protein binders with major interface residues exceeding 18 Å from all functional sites are considered deviating from the target functional region and are not considered priority candidates. Simultaneously, spatial conflict constraints are set in the system. ,in, This represents the minimum distance between any non-bonded atoms between the candidate protein conjugate and the target protein. If this distance is less than 2.0 Å, it indicates severe spatial overlap or atomic collisions.

[0165] The system combines the above constraints into a set of constraints for the generation of candidate protein conjugates and substitutes them into the data:

[0166] ;

[0167] For candidate protein conjugates Based on the threshold constraints mentioned above, candidate protein conjugates can be determined. It satisfies the EGFR functional site blocking constraint.

[0168] Step 2: Generate candidate protein conjugates based on the functional site blocking constraints. Divide interface residues and non-interface residues according to their spatial contact relationship with the target protein. While keeping the key residues involved in the functional site contact in the interface residues unchanged, perform sequence optimization on the non-interface region.

[0169] Generate candidate protein conjugates:

[0170] The system generates a set of candidate protein conjugates. The system is set to generate an initial quantity of [number]. Each candidate protein conjugate is represented as ,in, The amino acid sequence of the candidate protein-conjugate; For the backbone of candidate protein conjugates; This represents the initial binding posture of the candidate protein conjugate to EGFR. This is the set of predicted interface residues for candidate protein binders.

[0171] The system designs candidate protein conjugates as mini proteins or artificially designed scaffold proteins, and sets the generation parameters:

[0172] ;

[0173] in, This indicates that the length of the candidate protein conjugate is controlled between 55 and 75 amino acid residues; This indicates that the candidate form is either a mini protein or a scaffold protein; This indicates that the generative model has undergone 50 iterations. This indicates that the introduction of free cysteine ​​is prohibited in the initial design phase to reduce the risk of unintended disulfide bonds and aggregation. Therefore, the candidate generation input can be represented as: ;

[0174] in, To represent the candidate protein binder generation model, the system first in The candidate protein-binding backbone is generated nearby, positioning it close to the center of the EGFR functional site. .

[0175] Candidate protein conjugates The center of the skeleton is System requirements For example, candidate protein conjugates The center of the skeleton is In the aforementioned steps, the center of the EGFR hotspot region is ,but ,because ,therefore The skeleton position satisfies the generation constraint near the functional site.

[0176] The system selects a set of residues on the surface of the candidate protein conjugate that face the EGFR functional region as the initial interface residue set: ;

[0177] In this embodiment, The length is 64aa, and the predicted interface residues are:

[0178] ;

[0179] Right now There are 12 interface residues in total.

[0180] The system then assigns amino acids to these interface residues that tend to form hydrogen bonds, salt bridges, hydrophobic contacts, or shape complementarity. After the backbone is determined, the system generates an amino acid sequence for each candidate backbone. Candidate sequences can be represented as:

[0181] ;

[0182] For example, the system is a candidate protein conjugate. The generated example sequence is:

[0183] RAFGNGEMQDLFK;

[0184] The sequence length is ,satisfy ,therefore, The candidate length constraint is satisfied.

[0185] To ensure that candidate protein conjugates preferentially attach to key functional regions of EGFR, the system allows setting a candidate generation loss function:

[0186] ;

[0187] In this embodiment, the weight is set as follows: For candidate protein conjugates The system knows that: , , , , ;

[0188] Substitute into the total loss function: ;

[0189] because Lower, indicating The system adequately met the EGFR functional site blocking constraints during the generation phase. The system then performed constraint screening on 1000 initial candidate protein conjugates. Screening conditions could be set as follows: , , , .

[0190] The initial generation results are shown in Table 4:

[0191] Table 4

[0192]

[0193] The system outputs candidate protein conjugates that satisfy the initial functional blocking constraints, and each candidate protein conjugate includes at least:

[0194] ;

[0195] Perform sequence optimization for non-interface regions:

[0196] Let the candidate protein conjugate be The set of all its residues is represented as If the residues in the candidate protein binder With any residue in EGFR The distance is not greater than the interface distance threshold. Then Determined to be an interface residue, already set Candidate protein conjugates The set of interface residues is:

[0197] ;

[0198] Therefore, the set of non-interface residues is ,Right now In addition to the 12 interface residues mentioned above, the other 52 residues belong to non-interface regions.

[0199] Combination Figure 4 The system first identifies the set of key interface residues from the interface residues. These residues are involved in contacting hotspot residues such as K721, E738, T790, and G719 in EGFR, and therefore remain unchanged during sequence optimization.

[0200] The key residue fixation constraint of the interface is represented as ,in, Indicates the first (unoptimized) Amino acid type, Indicates the optimized first... The system requires that the optimized candidate protein conjugate still meet the functional blocking constraints of the previous step and restricts interfacial conformational changes. ;

[0201] The system performs sequence redesign on non-interface residues, and the optimization objective can be expressed as:

[0202] ;

[0203] In this embodiment, the weight is set to... The system judges: Although The system meets the functional site coverage requirements, but its solubility score of 0.68 is low and its aggregation risk of 0.32 is relatively high, therefore optimization of non-interface regions is needed. The system only optimizes non-interface residue sets. The system selects replaceable sites without altering key interface residues. For example, the system proposes non-interface residue replacement schemes as shown in Table 5:

[0204] Table 5

[0205]

[0206] After optimization, the system re-predicts. The stability, solubility, expression feasibility, and risk indicators were obtained, as shown in Table 6:

[0207] Table 6

[0208]

[0209] As you can see, the optimized interface This indicates that the interface design remains largely stable. (Optimized weighted coverage) Number of hotspot residues Therefore, non-interface optimization did not break... Blocking ability of EGFR functional sites.

[0210] The overall optimization target value before optimization is:

[0211] ;

[0212] The optimized overall target value is:

[0213] ;

[0214] Optimization and improvement amount This indicates that sequence optimization in non-interface regions makes... The overall stability, solubility, and developability were significantly improved without significantly reducing its coverage of EGFR functional sites. Finally, the system output the optimized candidate protein conjugate:

[0215] .

[0216] Step 3: For the candidate protein conjugates obtained in Step 2, predict the structure of the complexes they form with the target protein, and calculate the function blocking score based on the complex structure; the function blocking score is used to quantify the blocking ability of the candidate protein conjugates to the key functional regions of the target protein.

[0217] Predicting the structure of the candidate protein conjugate complex with the target protein:

[0218] The system will optimize the protein conjugate The structure of the binder-target complex is predicted by inputting the EGFR target protein structure into the complex structure prediction module.

[0219] The input for this step is: ;

[0220] The system Multiple complex structure predictions were performed with EGFR.

[0221] For example, in this embodiment, five independent predictions are made for the same candidate. The system determines whether candidate protein conjugates are stably bound to the target functional region of EGFR. Let the center of the EGFR hotspot region be... Candidate protein conjugates The interface center in the complex is:

[0222] ;

[0223] The distance between the center of the candidate conjugate interface and the center of the EGFR hotspot region is:

[0224] ;

[0225] Since this distance is within the set attitude constraint range Therefore, the system judges The binding posture is located near the EGFR functional site, which satisfies the complex posture requirements.

[0226] The system identifies interface residues between the candidate protein conjugate and EGFR residues in the complex structure. If the candidate protein conjugate residues... With EGFR residues The minimum distance satisfies If so, then it is considered that the two have formed an interface contact. (Already set) The set of side interface residues for candidate protein-binding compounds is as follows:

[0227] ;

[0228] The set of EGFR side interface residues is: ;

[0229] in, These are core hotspot residues. These residues belong to the auxiliary functional region.

[0230] The system further constructs a residue contact map between the candidate protein conjugate and EGFR:

[0231] ;

[0232] in, Indicates candidate protein-binding residues With EGFR residues Forming contact, This indicates that no contact has been formed. The total number of interface contacts is represented as: ;

[0233] In this embodiment, candidate protein conjugates were obtained through systematic statistical analysis. There are a total of 38 residue contacts with EGFR. The contact details for core hotspot residues of EGFR are shown in Table 7.

[0234] Table 7

[0235]

[0236] Therefore, the optimized candidate protein conjugates Five EGFR core hotspot residues were accessed.

[0237] The system calculates the interfacial area based on the change in solvent-accessible surface area before and after the formation of the complex. The solvent-accessible surface area of ​​the candidate protein conjugate in its free state; The solvent-accessible surface area of ​​EGFR in its free state; This represents the total solvent-accessible surface area in the complex state.

[0238] The buried surface area can then be expressed as: ;

[0239] In this embodiment, the prediction is obtained , , .therefore ;

[0240] If the interface area is defined as the average buried area on one side, then: ;

[0241] therefore, The predicted interface area with EGFR is This value indicates that a relatively sufficient binding interface was formed between the candidate protein conjugate and EGFR.

[0242] The confidence level of the overall structure of the complex is expressed as: Interface credibility is represented as The system received ,in, This indicates that the overall structure prediction of the complex is relatively reliable. This indicates that the interface prediction between the candidate protein conjugate and EGFR has high reliability.

[0243] To avoid the possibility of a single prediction yielding a favorable pose by chance, the system performs a consistency analysis on the results of five predictions. For reference, calculate the interface RMSD for the remaining poses: ;

[0244] In this embodiment, the interface RMSD results of the 5 predictions are shown in Table 8:

[0245] Table 8

[0246]

[0247] The average RMSD of the interface is: ;

[0248] The system can define attitude consistency as: ;

[0249] If a normalization method is used, this value can also be mapped to... System records This indicates that multiple predictions were made. All bound proteins were stably localized near the EGFR functional site, exhibiting good consistency in binding posture. The system further evaluated the degree of shape complementarity between the candidate protein binders and the EGFR functional region. Geometric complementarity can be expressed as:

[0250] ;

[0251] in, Scoring is based on the complementary shapes of the interfaces; Score for the reasonableness of the contact distance; To determine the reasonableness score of the combined direction, in this embodiment, the system calculates... .

[0252] but The geometric complementarity with the EGFR functional site region is as follows:

[0253] ;

[0254] This indicates that the two structures have a good fit in terms of spatial shape. The system also needs to determine whether there are severe atomic collisions in the composite structure. Let... This represents the minimum distance between the candidate protein conjugate and any non-bonding atom of EGFR. In this embodiment, the predicted distance is... Because the minimum spatial conflict distance is set to ,and Therefore, the system determines that the structure of the complex does not have a serious spatial conflict.

[0255] To quantify the degree of spatial conflict, we can define: ;

[0256] Substitute: This indicates that the structural conflict penalty of the complex is 0.

[0257] The system summarizes the predicted results of the complex structure as shown in Table 9:

[0258] Table 9

[0259]

[0260] Therefore, the system determines It can form a stable binder-target complex with EGFR, and its binding interface is located near the functionally critical region of EGFR, which can be used to proceed to the calculation of function blocking score.

[0261] Final output:

[0262] Calculating Functional Blockade Score: After obtaining the structure of the complex of the candidate protein conjugate and the target protein, the system further proceeds to the step of calculating the functional blockade score. For example...Figure 5 As shown, the system extracts residue contact maps based on the complex interface and further calculates scoring features such as functional site coverage, number of hotspot residue contacts, degree of spatial occlusion, and interface area to evaluate whether candidate protein conjugates truly cover and block the key functional regions of the target protein.

[0263] The system further calculates the blocking ability of candidate protein conjugates against key functional regions of EGFR based on the structure of the complex. Input for this step: ;

[0264] in, This represents the set of residues that are actually contacted on the EGFR side. This represents the actual set of interface residues on the candidate protein-binding side. Residue contact diagram, Represents the interface area. This indicates the overall confidence level of the complex. Indicates the credibility of the interface. This indicates consistency in attitude predictions across multiple predictions. This indicates the degree of spatial conflict. The contact details of the core hotspot residues in EGFR are shown in Table 10:

[0265] Table 10

[0266]

[0267] Therefore, the standard coverage rate is substituted into the calculation: ;

[0268] illustrate It covers all five core hotspot residues of EGFR.

[0269] Since different hotspot residues have varying degrees of influence on EGFR function, the system further calculates a weighted coverage ratio. Let the weight of each hotspot residue be... .

[0270] Weighted coverage ratio is used for calculation: ;

[0271] The system further calculates the contact score of hotspot residues. The calculation formula is as follows:

[0272] ;

[0273] in: ;

[0274] because It forms contacts within 6.0 Å with K721, E738, T790, L858, and G719, therefore It can also record the number of hotspot residues in contact. ,illustrate Five core hotspot residues were contacted, meeting the minimum number of hotspot contacts required in the aforementioned steps. .

[0275] Simply contacting hotspot residues is insufficient; the system also needs to determine whether candidate protein binders spatially obstruct the original functional region of EGFR. For example, for the EGFR kinase region, the ATP-binding pocket, substrate entry region, or disease-related activation region can be considered as the original functional space. Let the original functional binding space be... The overlap volume between the candidate protein binder and this functional space is Spatial occlusion score is defined as: ;

[0276] The system predicts based on the complex structure. ,therefore, , The original functional space of EGFR was obstructed by approximately 71.4%.

[0277] The system further calculates the interface area score based on the obtained interface area. The interface area predicted in the preceding steps... The system sets the reference value for the desired interface area. The interface area score is defined as:

[0278] ;

[0279] illustrate It forms a binding interface with EGFR that is close to the expected size.

[0280] System definition combined with attitude stability score:

[0281] ;

[0282] in, The overall confidence level of the complex; To improve the credibility of the interface; To ensure consistency in pose predictions across multiple scenarios; To normalize the degree of spatial conflict.

[0283] In this embodiment: ;

[0284] Set weights ,

[0285] but ,therefore The combined attitude stability score is .

[0286] Functional blocking scoring function:

[0287] ;

[0288] Set the weight to ,and , , , , .

[0289] Substitute: ;

[0290] The system allows setting function blocking rating levels:

[0291] ;

[0292] because Therefore, the system will The candidate protein conjugate was identified as having high-function blocking ability. The system can also compare the functional blocking scores of multiple candidate protein conjugates. For example, Table 11:

[0293] Table 11

[0294]

[0295] This shows that and These candidates perform well in terms of functional site coverage, hotspot residue contact, and spatial occlusion, making them more suitable as priority candidates for subsequent off-target risk assessment and multi-target screening.

[0296] Step 4: Construct a set of non-target proteins, predict the cross-binding ability of the candidate protein conjugates obtained in Step 2 with the non-target proteins, and calculate the off-target risk score by combining the sequence homology and structural similarity between the non-target proteins and the target proteins, as well as the expression risk of the non-target proteins in normal tissues.

[0297] Off-target risk assessment: The system first identifies the set of non-target proteins requiring off-target risk assessment based on sequence homology, structural similarity, similar surface pockets, and normal tissue expression. .

[0298] EGFR belongs to the ERBB receptor family; therefore, the system preferentially selects proteins with some homology or structural similarity to EGFR as non-target proteins, such as... The set of non-target proteins can be represented as: .

[0299] In this embodiment, the following settings are provided: In other words, a protein can be included in the non-target protein set if any of the indicators related to its sequence, structure, pocket, or normal tissue expression risk reaches a threshold. The system then predicts the candidate protein binding compounds. With each non-target protein Potential complex structures and cross-binding capabilities were calculated. Cross-association ability can be expressed as: ;

[0300] in, The score represents the area of ​​the interface between the candidate conjugate and the non-target protein. Confidence level of the cross-linked complex structure; Score based on the number of contacts at the cross interface; The score is given for the geometric complementarity of the cross interface.

[0301] In this embodiment, weights are set. ,for and The system obtains cross-combination prediction. ,

[0302] but: ;

[0303] illustrate It has some potential cross-binding ability to ERBB2.

[0304] For each non-target protein, there is a risk of off-target effects. :

[0305] ;

[0306] In this embodiment, the settings are as follows: ,for The system obtained:

[0307] ;

[0308] Substitute: ;

[0309] illustrate The off-target risk for ERBB2 is at a moderately high level, and penalties or warnings should be implemented in subsequent multi-target screening. The system... The same calculations were performed on several non-target proteins, and the results are shown in Table 12:

[0310] Table 12

[0311]

[0312] For example, for : ;

[0313] For multiple non-target proteins, the system can summarize the off-target risk score using the maximum value, weighted sum, average value, or quantile.

[0314] The maximum value method focuses on the most dangerous non-target proteins. This method is suitable for drug candidate screening scenarios with high safety requirements.

[0315] The weighted sum method considers the importance of different non-target proteins: ,in, Indicates the first Risk weights for non-target proteins.

[0316] In this embodiment, the following settings are made: .

[0317] ;

[0318] Using a biased maximum value method, A comprehensive risk-weighted method is adopted. ;

[0319] The system can be set to the following off-target risk levels:

[0320] ;

[0321] In this embodiment, the maximum value method is used. ,because Therefore, the system will This candidate has been identified as having a moderate off-target risk. It does not need to be immediately eliminated, but should be penalized for off-target risks in subsequent steps, and the main sources of risk should be indicated in the output. The main sources of off-target risk are ,because It has the highest risk value among all non-target proteins. The system can further indicate the reasons for ERBB2's higher risk, including... ,illustrate It has a certain predictive cross-combination ability with ERBB2; This indicates that ERBB2 and EGFR have a high degree of similarity in related structural regions; This indicates a high risk of expression of the non-target protein in normal tissues, therefore the risk of misbinding needs close monitoring. If the off-target risk of the candidate protein conjugate exceeds the preset optimization threshold, for example... The system can mark it as a candidate that needs optimization and propose residue-level modification suggestions.

[0322] For example, system discovery In , , Simultaneously involved in the prediction of EGFR and ERBB2 exposure, among which It forms effective contact with EGFR T790, but may also form cross-contact with similar pocket regions of ERBB2. The system can suggest... or The aim is to reduce non-specific binding to the conserved hydrophobic regions of ERBB2 while preserving as much blocking effect as possible on EGFR functional hotspot residues. However, the system needs to restrict modifications so as not to impair EGFR functional blocking ability; therefore, the optimized system still needs to meet certain requirements. .

[0323] The system can also compare the off-target risks of multiple candidates, as shown in Table 13:

[0324] Table 13

[0325]

[0326] Step 5: Based on the functional blocking score obtained in Step 3 and the off-target risk score obtained in Step 4, perform multi-target comprehensive screening by combining the target binding score, structural stability score and exploitability score of the candidate protein conjugate, determine the preferred candidate, and output the sequence information, structural information and various score reports of the preferred candidate protein conjugate.

[0327] Multi-target comprehensive screening: Target-based scoring This is used to evaluate whether candidate protein conjugates can form stable, reliable binding with good interfacial quality to the target protein EGFR.

[0328] System definition: ;

[0329] in, The predicted binding ability of candidate protein conjugates to EGFR; Score the interface area; To improve the credibility of the interface; The score is for geometric complementarity.

[0330] In this embodiment, the settings are as follows: ,for According to the aforementioned steps, , , , .

[0331] ;

[0332] illustrate It exhibits good targeting and binding ability with EGFR. Structural stability score. Used to evaluate whether candidate protein conjugates can fold stably and whether there are obvious spatial conflicts or unstable loop regions.

[0333] System definition: ;

[0334] in, Confidence level of the self-folding of candidate protein conjugates; This is a predicted value for folding stability; To normalize the degree of spatial conflict; This poses a risk to unstable ring zones.

[0335] In this embodiment, the settings are as follows: ,for have:

[0336] .

[0337] Substitute: ;

[0338] illustrate It has good structural stability.

[0339] Developability score Used to evaluate whether candidate protein conjugates have good solubility, expression feasibility, low aggregation risk, and low manufacturing risk.

[0340] System definition: ;

[0341] in, Risk associated with low-complexity sequences; To create risk.

[0342] In this embodiment, the settings are as follows: .

[0343] for After optimization, the result is ;

[0344] Substitution ,illustrate It has good solubility, expression feasibility and overall development potential.

[0345] System settings for multi-target filtering thresholds .

[0346] In this embodiment, the following settings are provided: ;

[0347] for Already exist: ,

[0348] therefore, It meets the basic screening criteria for entering the comprehensive ranking.

[0349] The system can first perform Pareto non-dominated ranking based on functional blocking score, target binding score, structural stability score, exploitability score, and off-target risk score to obtain a Pareto candidate set; then, within the Pareto candidate set, it can perform secondary ranking based on safety priority rules and comprehensive scores.

[0350] However, due to The system can mark it as a "candidate with moderate off-target risk" and lower its recommendation priority in subsequent rankings, or indicate that further optimization is needed. The system can also further calculate a comprehensive score. It is used to perform secondary sorting among candidates.

[0351] The overall score is expressed as follows: ;

[0352] In this embodiment, the settings are as follows: For candidate protein conjugates ,

[0353] Substitute:

[0354] ;

[0355] This result indicates It has a high overall score, but because its off-target risk score reaches 0.648, the system will not simply regard it as the highest priority candidate, but will combine it with other candidates for multi-target comparison.

[0356] The system performed the same calculations on multiple candidate protein conjugates, and the results are shown in Table 14:

[0357] Table 14

[0358]

[0359] The system can set the following candidate classification rules:

[0360] Primary candidate: These candidates have strong function blocking ability and low off-target risk, and should be given priority for experimental verification.

[0361] Secondary candidate: These candidates have strong functional blocking ability, but there is a certain risk of off-target effects. It is recommended to optimize the residues before verification.

[0362] Level 3 candidates: If the stability or exploitability indicators are slightly lower, such candidates are not the first choice, but can be used as backup optimization targets.

[0363] Candidates were eliminated: or or Or it may be predicted that the target functional site cannot be stably bound.

[0364] Final output:

[0365] The output results can be represented as Table 15:

[0366] Table 15

[0367]

[0368] Although Overall rating Slightly higher However, due to The risk of off-target effect is high; the system can prioritize safety rules to prevent this. It should be given higher priority in experimental verification.

[0369] Output candidate results: The system outputs the final design results of the candidate protein conjugates, including the amino acid sequence, three-dimensional structure, complex structure formed with EGFR, binding interface residues, target functional site coverage, hotspot residue contact, spatial occlusion, functional blocking score, target binding score, structural stability score, exploitability score, off-target risk score, comprehensive score, candidate classification, and experimental validation priority.

[0370] In this embodiment, the candidate protein conjugate Due to its functional blocking score of 0.870 and off-target risk score of 0.432, it was listed as a first-level candidate and assigned experimental validation priority I; candidate protein conjugate Although its functional blocking score reached 0.924, covering five core hotspot residues of EGFR (K721, E738, T790, L858, and G719), its off-target risk score was 0.648, with ERBB2 being the primary source of risk. Therefore, it was classified as a secondary candidate and assigned experimental validation priority II. (Candidate protein conjugate) Due to a functional blocking score of 0.722, it was classified as a Level III candidate; the off-target risk score exceeded 0.70. and functional blocking scores below 0.70 Candidates are listed as either eliminated or low-priority candidates. The system simultaneously outputs an explanation report detailing the reasons for the candidate's recommendation, downgrade, or elimination, and proposes modifications for interface residues that may pose off-target risks. This step allows users to intuitively obtain candidate protein conjugates suitable for experimental validation and their associated risk warnings, thereby improving the efficiency of subsequent experimental validation and candidate drug development.

Claims

1. A smart design method for protein conjugates that blocks target protein functional sites and assesses off-target risk, characterized in that, It includes the following steps: Step 1: Obtain disease-related target protein information, perform feature analysis on the target protein information, identify key functional regions, and construct functional site blocking constraints based on the key functional regions; Step 2: Generate candidate protein conjugates based on the functional site blocking constraints. Divide interface residues and non-interface residues according to their spatial contact relationship with the target protein. While keeping the key residues involved in the functional site contact in the interface residues unchanged, perform sequence optimization on the non-interface region. Step 3: For the candidate protein conjugates obtained in Step 2, predict the structure of the complexes they form with the target protein, and calculate the functional blocking score based on the complex structure; the functional blocking score is used to quantify the blocking ability of the candidate protein conjugates on the key functional regions of the target protein. Step 4: Construct a set of non-target proteins, predict the cross-binding ability of the candidate protein conjugates obtained in Step 2 with the non-target proteins, and calculate the off-target risk score by combining the sequence homology and structural similarity between the non-target proteins and the target proteins, as well as the expression risk of the non-target proteins in normal tissues. Step 5: Based on the functional blocking score obtained in Step 3 and the off-target risk score obtained in Step 4, perform multi-target comprehensive screening by combining the target binding score, structural stability score and exploitability score of the candidate protein conjugate, determine the preferred candidate, and output the sequence information, structural information and various score reports of the preferred candidate protein conjugate.

2. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 1, characterized in that, The disease-related target protein information mentioned in step 1 includes one or more of the following: target protein amino acid sequence, three-dimensional structure, PDB file, predicted structure, protein name, species origin, disease type, functional category, natural ligand, substrate, receptor, interacting protein, and known mutation site information; The feature analysis includes analyzing the spatial coordinates, secondary structure, surface accessibility, charge, hydrophobicity, and conservation characteristics of target protein residues to form a structured data object of the target protein.

3. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 1, characterized in that, The key functional regions described in step 1 include one or more of the following: receptor binding site, ligand binding site, protein-protein interaction interface, enzyme active site, substrate binding pocket, key signal transduction region, allosteric regulation region, and user-specified hotspot residues. The key functional regions are determined through one or more of the following methods: user specification, structural annotation parsing, extraction of known complex interfaces, pocket identification, conservation analysis, or hotspot residue prediction. Functional site set Represented as: ； Hotspot residue set Represented as: ； in, It represents the set of all residues of the target protein. Indicates the first One target protein residue, Indicates functional locus score, Indicates hotspot residue score, and These represent the functional site threshold and the hotspot residue threshold, respectively.

4. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 3, characterized in that, The hotspot residue score is calculated based on at least two of the following: residue conservation, interfacial contact frequency, mutation effect, and energy contribution, and is expressed as follows: ； in, Indicates the first Conservation score of each residue, Indicates the frequency of interface contact. This indicates that the mutation affects the score. Indicates the energy contribution score. , , , These are the weight parameters.

5. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 1, characterized in that, The specific method for constructing functional site blocking constraints based on the functionally critical regions as described in step 1 includes: Functional site blocking constraints are generated based on the spatial coordinates, surface accessibility, charge distribution, hydrophobicity, secondary structure environment, and spatial extent of residues in functionally critical regions. Functional site blocking constraints include one or more of the following: target contact residue set, preferred coverage region, exclusion region, binding direction constraint, spatial distance constraint, spatial occlusion constraint, interface shape complementarity constraint, and candidate binding posture restriction; Among them, priority coverage areas Represented as: ； in, Represents any point in three-dimensional space. Indicates hotspot residues spatial coordinates, Represents Euclidean distance. Indicates the blocking radius.

6. The intelligent design method for protein conjugates targeting the functional sites of target proteins and assessing off-target risks according to claim 5, characterized in that, The specific method for generating candidate protein conjugates based on the functional site blocking constraint in step 2 includes: The coverage constraints of candidate protein conjugates on functional sites are expressed as follows: ； in, Indicates candidate protein conjugates. Indicates the first in the candidate protein conjugate One residue, Indicates the contact distance threshold; Candidate protein conjugates are generated using structure prediction models, diffusion generation models, sequence design models, protein language models, and structure generation models. The generated candidate protein conjugates satisfy the coverage constraints. The generation of the candidate protein conjugate includes backbone generation, sequence generation, and binding posture generation; The set of candidate protein conjugates is represented as follows: ； in, Indicates the first One candidate protein conjugate, Indicates the number of candidates; The candidate generation loss function is expressed as: ； in, This indicates the distance penalty for candidate protein binders deviating from the functional site region. This indicates a penalty for insufficient coverage of functional sites. This indicates a penalty for insufficient contact with hotspot residues. Indicates spatial conflict penalty, This indicates a penalty for insufficient structural stability. to These are weight parameters; The candidate protein conjugates include antibody fragments, nanobodies, mini proteins, artificially designed binding proteins, or scaffold proteins.

7. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 1, characterized in that, Step 2 describes the specific method for distinguishing interface residues from non-interface residues based on their spatial contact relationship with the target protein, including: The set of interfacial residues of candidate protein-conjugates is represented as: ； The set of non-interface residues is represented as: ； in, Indicates target protein, This represents the set of all residues in a candidate protein-binding compound. This represents the set of interface residues of candidate protein-binding compounds. This represents the set of non-interfacial residues of candidate protein-conjugates. Indicates the interface distance threshold; Fixed constraints are set for key interface residues involved in the contact of functionally critical regions: ； in, Represents the set of key residues at the interface. Indicates the residue type before optimization. Indicates the optimized residue type; For sequence redesign of non-interface residues, the optimization objective is expressed as: ； in, Indicates the stability score. Indicates solubility score. Indicates the feasibility score. This indicates a risk of gathering. Indicates potential development risk. to These are the weight parameters.

8. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 1, characterized in that, Step 3, which involves predicting the complex structure of each candidate protein conjugate with the target protein and calculating the functional blocking score based on the complex structure, specifically includes: The complex structure prediction output includes one or more of the following: binding attitude, interface residues, residue contact map, number of interface contacts, complex structure confidence, interface confidence, geometric complementarity, interface area, attitude consistency, and degree of spatial conflict. The functional blocking score is calculated based on at least two of the following: functional site coverage, hotspot residue contact score, spatial occlusion score, interfacial area score, and binding posture stability score, and is expressed as follows: ； in, Indicates the first Functional blocking scores for candidate protein conjugates Indicates the weighted functional site coverage. Indicates the contact score of hotspot residues. Indicates spatial occlusion score. The score represents the interface area. This indicates that the attitude stability score is combined. to These are the weight parameters.

9. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 1, characterized in that, The set of non-target proteins mentioned in step 4 includes one or more of the following: proteins that have sequence homology with the target protein, proteins that have structural similarity with the target protein, proteins that have similar surface pockets or domains, and proteins expressed in normal human tissues. Step 4 involves considering the sequence homology and structural similarity between the non-target protein and the target protein, as well as the expression risk of the non-target protein in normal tissues, to calculate an off-target risk score, expressed as follows: ； in, Indicating candidate protein conjugates Regarding the first Non-target protein Off-target risk score, This indicates the predicted binding ability of candidate protein conjugates to non-target proteins. This indicates the sequence similarity between non-target proteins and target proteins. Indicating structural similarity, This indicates that a normal organization is expressing risk. to These are weight parameters; For multiple non-target proteins, the overall off-target risk score is obtained by maximizing the value or by weighted summation: ； or: ; in, Indicates the first Risk weights for non-target proteins.

10. The intelligent design method for protein conjugates targeting functional sites of target proteins and assessing off-target risks according to claim 1, characterized in that, Step 5 describes a multi-target comprehensive screening method to determine the preferred candidates based on functional blocking scores and off-target risk scores, combined with the target binding scores, structural stability scores, and exploitability scores of the candidate protein conjugates. The specific method is as follows: The overall score is expressed as follows: ； in, Indicates the first A comprehensive score for each candidate protein conjugate Indicates the functional blocking score. Indicates the target binding score, Indicates the structural stability score. Indicates the potential for development score. Indicates the off-target risk score. to These are weight parameters; Candidate protein conjugates are sorted, graded, or eliminated based on the comprehensive score and a preset threshold, wherein candidate protein conjugates with off-target risk scores exceeding the preset threshold are downgraded or eliminated. Step 5 outputs the sequence information, structural information, and various scoring reports of the preferred candidate protein conjugates, which specifically include: The amino acid sequence, three-dimensional structure, complex structure, binding interface residues, functional site coverage, hotspot residue contact, spatial occlusion, various scores, candidate classification, experimental validation priority, and interpretation report of the candidate protein conjugate.

Images