Protein design method, device, equipment, medium and program product

Abstract

The invention discloses a protein design method, device and equipment, a medium and a program product, and is applied to the field of artificial intelligence. The method comprises the following steps: acquiring a protein framework, and extracting protein characteristics of the protein framework; obtaining a target protein sequence, and extracting target protein characteristics of the target protein sequence; fusing the protein features with the target protein features to obtain fusion features; the fusion feature is used for representing sequence information and structure information under interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence; and performing protein sequence recovery on the protein framework based on the fusion feature, recovering a predicted protein sequence corresponding to the protein framework, and performing protein structure prediction on the predicted protein sequence to obtain a predicted protein structure corresponding to the predicted protein sequence.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a protein design method, device, equipment, medium and program product. Background Art

[0002] Protein is an important macromolecular organic compound in living organisms. Protein design includes sequence design and structural design, and is widely used in medicine and drug development, biotechnology and industrial applications, materials science, diagnosis and treatment, and other fields.

[0003] In the related art, a protein design model is trained based on sequence information corresponding to a large amount of proteins, and a protein design model is fine-tuned based on a specific type of target protein to generate a protein that binds to the specific type of target protein. For example, a target protein is an antigen (Antigen, Ag), which refers to a substance that can cause antibody production, such as bacteria, viruses, pollen, etc. A protein is an antibody (Antibody, Ab), which refers to a protein with a protective effect produced due to the stimulation of an antigen.

[0004] However, related technologies have difficulty in capturing the interactions between proteins and target proteins, resulting in inaccurate protein design. Summary of the invention

[0005] The present application provides a protein design method, device, equipment, medium and program product. The technical solution is as follows:

[0006] In one aspect, a method for protein design is provided, the method comprising:

[0007] Acquiring a protein framework, and extracting protein features of the protein framework;

[0008] Obtaining a target protein sequence, and extracting target protein features of the target protein sequence;

[0009] The protein feature and the target protein feature are fused to obtain a fusion feature; the fusion feature is used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence;

[0010] Based on the fusion feature, protein sequence recovery is performed on the protein framework to recover the predicted protein sequence corresponding to the protein framework, and protein structure prediction is performed on the predicted protein sequence to obtain the predicted protein structure corresponding to the predicted protein sequence.

[0011] In some embodiments, the sample protein features include: sample protein sequence features, sample protein pairing features and sample protein structure, the sample target protein features include: sample target protein MSA features, sample target protein pairing features and sample target protein structure, and the sample fusion features include: sample fusion sequence features and sample fusion pairing features;

[0012] The fusing the sample protein feature with the sample target protein feature to obtain the sample fusion feature comprises:

[0013] Fusion of the sample protein sequence feature and the sample target protein MSA feature to obtain the sample fusion sequence feature;

[0014] Based on the sample protein structure and the sample target protein structure, the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature are fused to obtain the sample fusion pairing feature.

[0015] In some embodiments, the fusing the sample protein sequence feature and the sample target protein MSA feature to obtain the sample fusion sequence feature comprises:

[0016] Performing a linear transformation on the sample protein sequence features to obtain a transformed sample protein sequence feature; and performing a linear transformation on the first row of sample target protein MSA features in the sample target protein MSA features to obtain a transformed first row of sample target protein MSA features;

[0017] Based on cross attention, the transformed sample protein sequence features and the transformed first-row sample target protein MSA features are fused to obtain the sample fusion sequence features.

[0018] In some embodiments, based on the sample protein structure and the sample target protein structure, the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature are fused to obtain the sample fusion pairing feature, including:

[0019] Determining the protein coordinates corresponding to each amino acid residue in the sample protein structure based on the carbon atom coordinates of each amino acid residue in the sample protein structure; and determining the target protein coordinates corresponding to each amino acid residue in the sample target protein structure based on the carbon atom coordinates of each amino acid residue in the sample target protein structure;

[0020] Calculating the protein pairing distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample protein structure; and calculating the target protein pairing distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample target protein structure;

[0021] Based on the protein pairing distance and the target protein pairing distance, the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature are fused to obtain the sample fusion pairing feature.

[0022] In some embodiments, the method of obtaining the sample fusion pairing feature by fusing the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature based on the protein pairing distance and the target protein pairing distance includes:

[0023] Mapping is performed based on the sample protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature; and mapping is performed based on the sample target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature;

[0024] Calculating the outer product average based on the sample protein sequence feature and the sample target protein MSA feature to obtain the third part of the fusion pairing feature; and determining the transposition of the third part of the fusion pairing feature as the fourth part of the fusion pairing feature;

[0025] The first part of the fusion pairing features, the second part of the fusion pairing features, the third part of the fusion pairing features and the fourth part of the fusion pairing features are spliced ​​to obtain the sample fusion pairing features.

[0026] In some embodiments, the performing of mapping based on the sample protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature includes:

[0027] Performing a linear transformation on the sample protein pairing feature to obtain a transformed sample protein pairing feature; and performing a linear transformation on the protein pairing distance to obtain a transformed protein pairing distance;

[0028] The transformed sample protein pairing feature and the transformed protein pairing distance are added to obtain the first part of the fusion pairing feature.

[0029] In some embodiments, the step of performing mapping based on the sample target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature includes:

[0030] Performing a linear transformation on the sample target protein pairing feature to obtain a transformed sample target protein pairing feature; and performing a linear transformation on the target protein pairing distance to obtain a transformed target protein pairing distance;

[0031] The transformed sample target protein pairing feature and the transformed target protein pairing distance are added to obtain the second part of the fusion pairing feature.

[0032] In some embodiments, the calculating the outer product average based on the sample protein sequence feature and the sample target protein MSA feature to obtain the third part of the fusion pairing feature includes:

[0033] The sample protein sequence features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized protein sequence features; and the sample target protein MSA features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized target protein MSA features;

[0034] The outer product average of each element in the transformed normalized protein sequence feature and each element in the transformed normalized target protein MSA feature is calculated, and the outer product average is linearly transformed to obtain the third part fusion pairing feature.

[0035] In some embodiments, the method further comprises:

[0036] In the presence of a sample interchain feature or a sample interaction feature between the sample protein and the sample target protein, a sample updated sequence feature and a sample updated pairing feature are obtained based on the sample interchain feature or the sample interaction feature, and the sample protein structure and the sample target protein structure;

[0037] The sample fusion sequence feature is updated based on the sample update sequence feature to obtain an updated sample fusion sequence feature; and the sample fusion pairing feature is updated based on the sample update pairing feature to obtain an updated sample fusion pairing feature.

[0038] In some embodiments, in the presence of a sample interchain feature or a sample interaction feature between the sample protein and the sample target protein, obtaining a sample updated sequence feature and a sample updated pairing feature based on the sample interchain feature or the sample interaction feature, as well as the sample protein structure and the sample target protein structure, comprises:

[0039] In the case where the sample inter-chain feature exists, the sample interaction feature is determined based on the sample inter-chain feature and the inter-chain connection relationship; or, in the case where the sample interaction feature exists, the sample inter-chain feature is determined based on the sample interaction feature and the inter-chain pseudo connection relationship;

[0040] Performing a linear transformation on the sample interaction feature to obtain the sample update sequence feature; and performing a linear transformation on the sample inter-chain feature to obtain the sample update pairing feature.

[0041] In another aspect, a protein design apparatus is provided, the apparatus comprising:

[0042] An acquisition module, used for acquiring a protein framework and extracting protein features of the protein framework;

[0043] The acquisition module is used to acquire the target protein sequence and extract the target protein features of the target protein sequence;

[0044] A fusion module, used to fuse the protein feature with the target protein feature to obtain a fusion feature; the fusion feature is used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence;

[0045] A processing module is used to perform protein sequence recovery on the protein framework based on the fusion feature to recover the predicted protein sequence corresponding to the protein framework, and to perform protein structure prediction on the predicted protein sequence to obtain the predicted protein structure corresponding to the predicted protein sequence.

[0046] On the other hand, a computer device is provided, comprising: a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the protein design method as described above.

[0047] On the other hand, a computer-readable storage medium is provided, wherein the computer-readable storage medium stores a computer program, and the computer program is loaded and executed by a processor to implement the protein design method as described above.

[0048] On the other hand, a computer program product is provided, which includes computer instructions stored in a computer-readable storage medium, and a processor obtains the computer instructions from the computer-readable storage medium, so that the processor loads and executes them to implement the protein design method as described above.

[0049] The beneficial effects brought by the technical solution provided by the embodiment of the present application include at least:

[0050] The invention provides a collaborative design method of protein sequence and protein structure, which can utilize a pre-constructed protein framework and, based on a specified target protein sequence, fully express the sequence information and structural information of the target protein indicated by the target protein sequence and the interaction between the target protein and the protein corresponding to the protein framework by fusing the protein features of the protein framework and the target protein features of the target protein sequence, and design a protein with the ability to bind to and interact with the target protein, thereby improving the accuracy of protein design and ensuring the rationality of the protein structure while designing the protein sequence. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0052] Figure 1 is a structural block diagram of a computer system provided by an exemplary embodiment of the present application;

[0053] Figure 2 is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application;

[0054] Figure 3 is a schematic block diagram of a protein design model provided by an exemplary embodiment of the present application;

[0055] Figure 4 is a schematic block diagram of a protein design model provided by an exemplary embodiment of the present application;

[0056] Figure 5 is a flow chart of a protein design method provided by an exemplary embodiment of the present application;

[0057] Figure 6 is a schematic block diagram of a protein design model provided by an exemplary embodiment of the present application;

[0058] Figure 7 is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application;

[0059] Figure 8 is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application;

[0060] Fig. 9 is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application;

[0061] Fig.10 is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application;

[0062] Fig.11 is a distribution diagram of three-cycle ipTM scores provided by an exemplary embodiment of the present application;

[0063] Fig.12 is a block diagram of a protein design apparatus provided by an exemplary embodiment of the present application;

[0064] Fig.13 It is a structural block diagram of a computer device provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below with reference to the accompanying drawings.

[0066] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application. Instead, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0067] The terms used in this application are for the purpose of describing specific embodiments only and are not intended to limit this application. The singular forms of "a", "said" and "the" used in this application and the appended claims are also intended to include plural forms unless the context clearly indicates other meanings. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more associated listed items.

[0068] It should be understood that, although the terms first, second, etc. may be used in the present application to describe various information, these information should not be limited to these terms. These terms are only used to distinguish the same type of information from each other. For example, without departing from the scope of the present application, the first parameter may also be referred to as the second parameter, and similarly, the second parameter may also be referred to as the first parameter. Depending on the context, the word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining".

[0069] It should be noted that before collecting relevant data of users (for example, proteins, target proteins, antigens, antibodies, genes, etc.) and during the process of collecting relevant data of users, this application can display a prompt interface, pop-up window or output voice prompt information, and the prompt interface, pop-up window or voice prompt information is used to prompt the user that its relevant data is currently being collected, so that this application only starts to execute the relevant steps of obtaining relevant data of users after obtaining the confirmation operation issued by the user to the prompt interface or pop-up window, otherwise (that is, when the confirmation operation issued by the user to the prompt interface or pop-up window is not obtained), the relevant steps of obtaining relevant data of users are terminated, that is, the relevant data of users are not obtained. In other words, all user data collected by this application are collected with the consent and authorization of the user, and the collection, use and processing of relevant user data need to comply with the relevant laws, regulations and standards of relevant countries and regions.

[0070] First, the nouns involved in the embodiments of the present application are briefly introduced:

[0071] Protein: Also known as protein molecule, it is an important macromolecular organic compound in organisms, mainly including carbon (C), hydrogen (H), oxygen (O), nitrogen (N) and other elements. Protein is composed of m amino acids and n peptide chains, and each protein has an independent and definite amino acid sequence. Among them, amino acids are the basic building blocks of proteins. The arrangement order of amino acids in proteins and the three-dimensional structure formed thereby constitute the diversity of protein structure. Proteins have primary structure, secondary structure, tertiary structure and quaternary structure. The arrangement order of amino acid residues in the peptide chain of proteins is called the primary structure of proteins. Among them, a peptide chain containing multiple peptide bonds is formed by connecting multiple amino acids to each other. Since some groups of amino acids participate in the formation of peptide bonds and lose a molecule of water, the amino acid unit in the polypeptide is called an amino acid residue, which can also be referred to as a residue.

[0072] Target protein: a protein located inside or on the cell surface that can be effectively acted upon by drugs or small molecule compounds and is used in the fields of protein research, drug research, etc. For example, a target protein can be at least one of an antigen, a viral target protein, and a receptor protein.

[0073] In the present application, the target protein is an antigen (Antigen, Ag), which refers to a substance that can cause antibody production, such as bacteria, viruses, pollen, etc. The protein is an antibody (Antibody, Ab), which refers to a protein with a protective effect produced due to the stimulation of an antigen.

[0074] Complementarity Determining Region (CDR): It is a part of the antibody that has high variability and specificity and can recognize and bind to a variety of different antigens. The CDR is located in the V region of the antibody. The light chain and heavy chain each have three CDRs, namely: CDR1, CDR2, and CDR3. Among them, CDR3 has the highest degree of variation.

[0075] In the embodiment of the present application, the protein framework is obtained by masking the amino acid sequence in the complementarity determining region of the protein sequence, and protein sequence recovery is performed on the protein framework, that is, the amino acid sequence in these complementarity determining regions is recovered.

[0076] Protein language model: It is a type of large language model based on deep learning, which is dedicated to processing and analyzing protein sequences. The working principle of protein language model is similar to that of natural language model. Its core idea is to regard protein sequence as a language, amino acid as a character, and learn the rules of the language by autoregression.

[0077] In an embodiment of the present application, the protein language model can be: a large language model based on the Transformer architecture, for example, an ESM (Evolutionary Scale Modeling) model. The input of the ESM model is a protein sequence, and the output is a feature representation, which includes at least one information such as protein structure and function, which can be further mapped or used in downstream models to achieve various predictions and analyses of proteins.

[0078] Invariant Point Attention Mechanism: It is a method for processing invariant features with invariant properties in the input data. It can utilize the characteristics of rotation and translation invariance, which means that the protein structure may actually be in any spatial position in three-dimensional space. No matter how the protein structure is rotated or translated in three-dimensional space, the attention weight calculated by the invariant point attention mechanism remains consistent.

[0079] Protein-Protein Interaction (PPI): refers to the process in which two or more protein molecules form a protein complex through non-covalent bonds, which can reveal information such as synergy, signal transduction and biological response between proteins.

[0080] Protein structure prediction unit: Specifically includes the Evoformer unit, which is used to process and optimize protein sequence and protein structure information. The Evoformer unit takes multiple sequence alignment (MSA) data as input, which includes protein sequences and their related homologous sequences. Through these homologous sequences, the Evoformer unit can capture the structural and function-related features of proteins that have been preserved during evolution.

[0081] Confidence score: It is an indicator score used to evaluate the accuracy and reliability of predicted protein sequence, predicted protein structure, and predicted complex structure, including: prediction template modeling score (pTM), interface prediction template modeling score (ipTM), local geometry quality prediction score (pLDDT), interface local geometry quality prediction score (ipLDDT), design score, and edit distance.

[0082] Figure 1 1 is a block diagram of a computer system provided by an exemplary embodiment of the present application. The computer system 100 can be implemented as a system architecture of a protein design method. The computer system 100 includes: a terminal 120 and a server 140.

[0083] The terminal 120 may be an electronic device such as a mobile phone, a tablet computer, a vehicle-mounted terminal (vehicle computer), a wearable device, a PC (Personal Computer), an unmanned reservation terminal, etc. The terminal 120 stores the protein design model, or a client that runs a target application may be installed in the terminal 120. The target application may provide a protein design function or a training function for a protein design model, which is not limited in the embodiment of the present application. In addition, the embodiment of the present application does not limit the form of the target application, including but not limited to App (Application) installed in the terminal 120, applets, etc., and may also be in the form of a web page.

[0084] Server 140 stores protein design models, or server 140 can provide protein design functions or protein model training functions, which is not limited in the present application embodiment. Server 140 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server, cloud database, cloud computing, cloud function, cloud storage, network service, cloud communication, middleware service, domain name service, security service, content distribution network (Content Delivery Network, CDN), and cloud server for basic cloud computing services such as big data and artificial intelligence platform that provide cloud computing services. Server 140 can be the background server of the above-mentioned target application, for providing background services for the client of the target application. In some optional embodiments, server 140 can also be implemented as a node in a blockchain system.

[0085] The terminal 120 and the server 140 may communicate with each other via a network, such as a wired or wireless network.

[0086] The protein design method provided in the embodiment of the present application can be performed by a computer device in each step. A computer device refers to an electronic device with data calculation, processing and storage capabilities. Figure 1 Taking the implementation environment of the scheme shown as an example, the protein design method can be executed by the terminal 120, for example, the client of the target application installed and run by the terminal 120 executes the protein design method, the protein design method can also be executed by the server 140, or the terminal 120 and the server 140 can interact and cooperate to execute it, and the embodiments of the present application are not limited to this.

[0087] Those skilled in the art will appreciate that the number of the above-mentioned terminals 120 may be more or less. For example, the above-mentioned terminal 120 may be only one, or may be dozens or hundreds, or more. The embodiment of the present application does not limit the number and device type of the terminal 120.

[0088] There is no clear solution for the de novo end-to-end protein design method for target proteins without any constraints and without providing reference sequence information.

[0089] In the related art, protein design includes two implementation methods:

[0090] 1. Train the protein design model based on the sequence information corresponding to a large number of proteins, and perform fine-tuning of the protein design model based on a specific type of target protein to generate a protein that binds to the target protein according to the specific type. For example, the target protein is an antigen (Antigen, Ag) and the protein is an antibody (Antibody, Ab). The protein design model can perform design on the complementary determining region of the antibody, especially the CDRH3 region. It performs pre-training based on a large number of antibodies, and performs fine-tuning based on a specific antigen to generate an antibody that binds to the specific antigen and determine the antibody sequence.

[0091] 2. Based on the structural information corresponding to the protein, a protein design model for the co-design of protein sequence and protein structure is trained. For example, an end-to-end protein design model based on an equivariant graph neural network can perform co-design on protein sequence and protein structure, and its training data is protein structure. For example, the target protein is an antigen and the protein is an antibody. The protein design model can perform co-design on the complementarity determining regions in the antibody structure and antibody sequence, especially the CDRH3 region.

[0092] However, the limitations of the two implementation methods of the related technology include at least:

[0093] 1. First, protein design models do not explicitly consider structural information. The binding of target proteins to proteins is often accompanied by conformational changes and dynamic interactions. These complex interactions are difficult to capture only through protein sequence information. Secondly, protein design models are prone to generate a large number of infeasible protein sequence combinations in the absence of structural information constraints. In addition, protein design models lack a screening mechanism based on structural information, making it difficult to effectively evaluate whether the designed protein has the expected function.

[0094] 2. First, the training data of the protein design model only relies on structural information, and has never considered pre-learning / transferring learning to a certain binding mode from a wide range of neutralizing protein sequences. Secondly, it is limited to designing the complementary determination region, and does not consider whether the framework region sequence is compatible. If the given framework region is not pre-experimented, it is difficult to select a reasonable framework region sequence.

[0095] The related art makes it difficult for protein design models to capture the interaction between proteins and target proteins, resulting in inaccurate protein design. The present application embodiment provides a method for the collaborative design of protein sequences and protein structures. By pre-constructing a protein framework library, using protein frameworks and target protein sequences, combined with the prediction of the amino acid sequence corresponding to the complementary determining region in the protein framework, the collaborative design of the entire protein sequence and protein structure is completed. Figure 2is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application. The protein design method is executed by a computer device, which may be Figure 1 The terminal 120 and / or the server 140 store the protein design model. The protein design method can be implemented based on the protein design model.

[0096] Step 1, obtain the protein framework 210, and extract the protein features 230 of the protein framework. The protein sequence in the protein framework 210 is represented as "H: QIQLVO...CARxxxxxxxWGQ...VSS; L: DVLMTQ...YKDDDK", "H" is the heavy chain, "L" is the light chain, and one letter in the letters "QILVOCARWGSDMTYK" represents a type of amino acid, and "xxx" is the masked amino acid sequence contained in the protein framework 210, that is, the amino acid sequence that needs to be restored. The protein features 230 include: protein sequence features, protein pairing features and protein structure. The term "feature" can also be referred to as "representation".

[0097] Step 2, obtaining a target protein sequence 220, and extracting a target protein feature 240 of the target protein sequence. The target protein includes an antigen, and the target protein sequence 220 is represented as "A: RVQPTE...", "A" represents an antigen, and one of the letters "RVQPTE" represents a type of amino acid. The target protein feature 240 includes: a target protein MSA feature, a target protein pairing feature, and a target protein structure.

[0098] Step 3, fusion protein feature 230 and target protein feature 240 are fused to obtain fusion feature 250; fusion feature 250 is used to characterize the sequence information and structural information under the interaction between the protein indicated by protein framework 210 and the target protein indicated by target protein sequence 220. Fusion feature 250 includes: fusion sequence feature and fusion pairing feature.

[0099] Step 4, based on the fusion feature 250, the protein framework 210 is restored to perform protein sequence recovery, the predicted protein sequence 260 corresponding to the protein framework 210 is recovered, and the predicted protein sequence 260 is predicted to perform protein structure prediction to obtain the predicted protein structure 270 corresponding to the predicted protein sequence 260. The protein sequence in the predicted protein sequence 260 is “H:QIQLVO…CARGYYGYYAFDYWGQ…VSS;L:DVLMTQ…YKDDDK”, GYYGYYAFD” is the masked amino acid sequence in the recovered protein framework 210. Optionally, complex structure prediction can be performed based on the fusion feature 250 to obtain a predicted complex structure, and the predicted protein structure 270 can also be separated from the predicted complex structure. Confidence scores 280 corresponding to the predicted protein sequence 260 and the predicted protein structure 270 can also be generated: ipTM=0.892, iplDDT=0.706, which are used to implement the screening of the predicted protein sequence 260 and the predicted protein structure 270.

[0100] In summary, the protein design method provided in the embodiment of the present application utilizes the protein framework and the specified target protein sequence. Since the protein and the target protein will interact and influence each other, the interaction between the protein and the target protein is fully represented by the fusion feature, thereby realizing the coordinated design of the protein sequence and protein structure and improving the accuracy of protein design. The following is a detailed example.

[0101] The protein design method provided in the embodiments of the present application can be implemented based on the protein design model. The following embodiments respectively illustrate the model structure, application stage and training stage of the protein design model.

[0102] Part I: Model structure of protein design models

[0103] Figure 3 It is a schematic block diagram of a protein design model provided by an exemplary embodiment of the present application.

[0104] Optionally, the model structure of the protein design model 200 for executing the protein design method includes: a target protein feature generation unit 20 and a protein feature generation unit 10 connected in parallel; a feature fusion unit 30 connected to both the target protein feature generation unit 20 and the protein feature generation unit 10; a complex structure prediction unit 40 and a protein sequence recovery unit 50 connected to the feature fusion unit 30, and a protein structure prediction unit 60 connected to the protein sequence recovery unit 50.

[0105] Among them, the target protein feature generation unit 20 is used to extract the target protein features of the target protein sequence, and is used to extract the sample target protein features of the sample target protein sequence; the protein feature generation unit 10 is used to extract the protein features of the protein framework, and is used to extract the sample protein features of the sample protein; the feature fusion unit 30 is used to fuse the protein features and the target protein features, and is used to fuse the sample protein features and the sample target protein features; the complex structure prediction unit 40 is used to perform complex structure prediction; the protein sequence recovery unit 50 is used to perform protein sequence recovery; the protein structure prediction unit 60 is used to perform protein structure prediction.

[0106] Figure 4 It is a schematic block diagram of a protein design model provided by an exemplary embodiment of the present application.

[0107] Optionally, the model structure of the protein design model 200 further includes: at least one of a feature recycling unit 70, a structure recycling unit 80, and a sequence recycling unit 90. The input end of the feature recycling unit 70 is connected to the output end of the target protein feature generation unit 20, and the output end of the feature recycling unit 70 is connected to the input end of the target protein feature generation unit 20. The feature recycling unit 70 is used to iteratively use the previously generated target protein features or sample target protein features to improve the accuracy of the target protein features or sample target protein features. The input end of the structure recycling unit 80 is connected to the output end of the complex structure prediction unit 40, and the output end of the structure recycling unit 80 is connected to the input end of the feature fusion unit 30. The structure recycling unit 80 is used to make the feature fusion unit 30 iteratively use the previously generated predicted complex structure to improve the accuracy of the fusion feature or the sample fusion feature. The input end of the sequence loop unit 90 is connected to the output end of the protein sequence recovery unit 50, and the output end of the sequence loop unit 90 is connected to the input end of the protein feature generation unit 10. The sequence loop unit 90 is used to iteratively perform masking on the candidate protein sequence to iteratively generate protein frameworks, and enable the protein feature generation unit 10 to use these iteratively generated protein frameworks to achieve iterative prediction of protein sequences and screening of protein sequences and protein frameworks.

[0108] In one possible implementation, the protein feature generation unit 10 in the protein design model 200 can be implemented as: a pre-trained language module (ESM-PPI) in series (Pretrained ESM-PPI Module), a pre-trained sequence-based protein structure prediction module (tFold-Ab) (Pretrained tFold-Ab Module). The target protein feature generation unit 20 can be implemented as: a pre-trained protein structure prediction module (Pretrained Protein Structure Prediction Module). The feature fusion unit 30 can also be called: a feature fusion module (Feature Fusion Module). The complex structure prediction unit 40 can also be called: a complex structure prediction module (Complex Prediction Module). The complex structure prediction unit 40 and the protein structure prediction unit 60 can be implemented as: Evoformer modules, respectively. The feature fusion unit 30, the complex structure prediction unit 40, the protein sequence recovery unit 50 and the protein structure prediction unit 60 can be combined into: an artificial intelligence-driven flexible docking module (AI-driven flexible Docking Module).

[0109] Part II: Application phase of protein design model

[0110] Figure 5 It is a flowchart of a protein design method provided by an exemplary embodiment of the present application.

[0111] The method is performed by a computer device, which may be Figure 1 The terminal 120 and / or the server 140 stores the protein design model. The method includes at least some of the steps 320, 340, 360, and 380:

[0112] Step 320, obtaining a protein framework and extracting protein features of the protein framework.

[0113] A protein sequence is a sequence composed of peptide chains and amino acids on the peptide chains in a protein. A protein sequence includes the amino acid sequence corresponding to the heavy chain and the amino acid sequence corresponding to the light chain. For example, a protein sequence is represented as "H: QIQLVO...; L: DVLMTQ...YKDDDK", "H" is the heavy chain, "L" is the light chain, and one of the letters "QILVOCARWGSDMTYK" is used to represent a type of amino acid.

[0114] The protein framework is also called a protein framework template or a protein framework region, which is obtained by masking at least a portion of the amino acid sequence in the entire protein sequence. The protein framework is crucial to the stability and solubility of the protein.

[0115] In some embodiments, the protein sequence includes a complementary determining region, and the protein framework is obtained by masking the amino acid sequence corresponding to the complementary determining region in the protein sequence. For example, the protein sequence in the protein framework is represented as "H: QIQLVO...CARxxxxxxxWGQ...VSS; L: DVLMTQ...YKDDDK", and "xxx" is the masked amino acid sequence contained in the protein framework, that is, the amino acid sequence that needs to be restored.

[0116] In some embodiments, a protein framework library is pre-constructed, and the protein framework library includes several protein frameworks. Exemplarily, a protein framework is obtained from a pre-constructed protein framework library, and the protein features of the protein framework are extracted. Optionally, a protein framework is obtained from the protein framework library, and the protein framework is the protein framework with the highest matching degree with one of the types, activities, and sources of the target protein. Or, at least a portion of the protein framework is obtained from the protein framework library, and at least a portion of the protein framework is a protein framework corresponding to at least one of the types, activities, and sources of the target protein. Wherein, the corresponding protein framework is set in advance based on the type, activity, and source of the target protein. Or, all protein frameworks are obtained from the protein framework library. When obtaining at least a portion or all of the protein frameworks, screening is also performed based on the confidence scores of the predicted protein sequences and predicted protein structures corresponding to each protein framework.

[0117] Protein features are features corresponding to the protein sequence indicated by the protein framework. In this embodiment, the protein features of the protein framework are extracted from at least two dimensions: sequence and structure. Optionally, protein features include: protein sequence features (Sequence Representation, s), protein pairing features (Pairwise Representation, z) and protein structure (x). Among them, the protein sequence features are used to indicate the sequence information of the protein framework. The protein pairing feature can also be called the protein paired feature, which is the feature corresponding to any two amino acid residues located in the same peptide chain, and is used to indicate the interaction between any two amino acid residues located in the same peptide chain. The protein structure is represented by the atomic coordinates of each amino acid residue in the protein, which is used to indicate the structural information of the protein. The atomic coordinates can be three-dimensional coordinates.

[0118] Step 340, obtaining a target protein sequence and extracting target protein features of the target protein sequence.

[0119] The target protein sequence is a sequence consisting of peptide chains in the target protein and amino acids on the peptide chains. For example, the target protein includes an antigen, and the target protein sequence is represented as "A: RVQPTE...", "A" represents the antigen, and one of the letters in "RVQPTE" represents a type of amino acid.

[0120] In some embodiments, a target protein sequence is obtained, a sequence database is searched based on the target protein sequence, at least one homologous sequence related to the target protein sequence is determined, the target protein sequence and at least one homologous sequence related to the target protein sequence are combined to obtain target protein multiple sequence alignment (MSA) data, and target protein features of the target protein MSA data are extracted. Optionally, when combining the target protein sequence and at least one homologous sequence related to the target protein sequence to obtain the target protein MSA data, the target protein sequence can be placed in the first row of the target protein MSA data.

[0121] Among them, the target protein MSA data includes: the target protein sequence and at least one homologous sequence related to the target protein sequence. Through these homologous sequences, the structural and functional related characteristics of the target protein retained during the evolution process can be captured. For example, the target protein includes an antigen, and the target protein sequence is represented as "A: RVQPTE...". After searching the sequence database, three homologous sequences are determined, which are represented as "RVVPSG...", "RVQPTV...", and "RVSPST...". The target protein sequence and the three homologous sequences are combined to obtain the target protein MSA data.

[0122] The target protein feature is the feature corresponding to the target protein sequence. In this embodiment, the target protein feature of the target protein sequence is extracted from at least the two dimensions of sequence and structure. Optionally, the target protein feature includes: target protein MSA feature (MSA Representation, m), target protein pairing feature (Pairwise Representation, z) and target protein structure (x). Among them, the target protein MSA feature is used to indicate the sequence information of the target protein sequence and its homologous sequence. The target protein pairing feature can also be called the target protein paired feature, which is the feature corresponding to any two amino acid residues located in the same peptide chain, and is used to indicate the interaction between any two amino acid residues located in the same peptide chain. The target protein structure is represented by the atomic coordinates of each amino acid residue in the target protein, which is used to indicate the structural information of the target protein. The atomic coordinates can be three-dimensional coordinates.

[0123] Step 360, the protein feature and the target protein feature are fused to obtain a fusion feature; the fusion feature is used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence.

[0124] Complex Representation is obtained by performing feature fusion on protein features and target protein features. Since there is interaction between protein and target protein, the two affect each other, so the fusion feature is used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence.

[0125] In this embodiment, the fusion feature includes: fusion sequence feature and fusion pairing feature. Among them, the fusion sequence feature is used to indicate the sequence information under the interaction between the protein and the target protein. The fusion pairing feature is used to indicate the information of the respective amino acid residues under the intrachain or interchain interaction between any two amino acid residues located in the same peptide chain of the protein and any two amino acid residues located in the same peptide chain of the target protein. The fusion pairing feature also includes the structural information and sequence information corresponding to the protein framework and the fusion protein sequence, which will be explained in subsequent embodiments.

[0126] Step 380, performing protein sequence recovery on the protein framework based on the fusion feature to recover the predicted protein sequence corresponding to the protein framework, and performing protein structure prediction on the predicted protein sequence to obtain the predicted protein structure corresponding to the predicted protein sequence.

[0127] Protein sequence recovery refers to predicting the amino acid sequence corresponding to the masked complementary determining region in the protein framework to restore the protein framework to a complete protein sequence, which is called predicted protein sequence. Protein structure prediction refers to the protein structure generated based on the predicted protein sequence, which is called predicted protein structure.

[0128] The fusion features include fusion sequence features and fusion pairing features. In some embodiments, protein sequence recovery is performed on the protein framework based on the fusion sequence features and the fusion pairing features, and the amino acid sequence corresponding to the masked complementary determination region in the protein framework is predicted to recover the predicted protein sequence corresponding to the protein framework. In addition, the predicted sequence features and predicted pairing features of the predicted protein sequence are extracted, and the predicted sequence features and the predicted pairing features are mapped through the invariant point attention mechanism to obtain the predicted protein structure.

[0129] In one possible implementation, the predicted sequence features and predicted pairing features are mapped through the invariant point attention mechanism, and the relative rotation angle and relative distance of each amino acid residue are predicted. Based on the relative rotation angle and translation information of each amino acid residue, the coordinates and angles of each atom are predicted, and the predicted protein structure is generated according to the coordinates and angles of each atom. In this embodiment, the use of the invariant point attention mechanism can map the predicted sequence features and predicted pairing features to a predicted atomic-level three-dimensional protein structure. It should also be noted that the extraction of predicted sequence features and predicted pairing features of the predicted protein sequence can be consistent with the method of extracting protein sequence features and protein pairing features of the protein framework.

[0130] In some embodiments, it also includes performing complex structure prediction based on the fusion feature to obtain a predicted complex structure. The predicted complex structure is a combination of the predicted protein structure and the updated target protein structure. Specifically, the fusion feature includes a fusion sequence feature and a fusion pairing feature. Through the invariant point attention mechanism, the fusion sequence feature and the fusion pairing feature are mapped to obtain a predicted complex structure. Optionally, the predicted protein structure can also be separated from the predicted complex structure. The protein design model provided in the embodiment of the present application can calculate the complex structure of the protein indicated by the protein framework and the target protein, and can also update the original target protein structure, and can also restore the predicted protein sequence corresponding to the protein framework, and can also generate a predicted protein structure corresponding to the predicted protein sequence.

[0131] In summary, the protein design method provided in the embodiment of the present application obtains a protein framework and extracts protein features of the protein framework; obtains a target protein sequence and extracts target protein features of the target protein sequence; fuses protein features with target protein features to obtain fusion features; fusion features are used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence; based on the fusion features, protein sequence recovery is performed on the protein framework to recover the predicted protein sequence corresponding to the protein framework, and protein structure prediction is performed on the predicted protein sequence to obtain the predicted protein structure corresponding to the predicted protein sequence. Accordingly, a collaborative design method for protein sequence and protein structure is provided, which can utilize a pre-constructed protein framework and, based on a specified target protein sequence, fully express the target protein indicated by the target protein sequence, and the sequence information and structural information under the interaction between the protein corresponding to the protein framework by fusing the protein features of the protein framework and the target protein features of the target protein sequence, and make full use of features of different dimensions and types to design proteins with the ability to bind to and interact with target proteins, thereby improving the accuracy of protein design and ensuring the rationality of protein structure while designing protein sequences.

[0132] In some embodiments, protein features include: protein sequence features, protein pairing features and protein structure, target protein features include: target protein MSA features, target protein pairing features and target protein structure, and fusion features include: fusion sequence features and fusion pairing features. Exemplarily, step 360 is specifically implemented as steps 362 and 364:

[0133] Step 362, fusion protein sequence features and target protein MSA features to obtain fusion sequence features.

[0134] Step 364, based on the protein structure and the target protein structure, the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature are fused to obtain a fusion pairing feature.

[0135] The feature fusion methods include: at least one of addition, multiplication, concatenation, linear combination, weighted summation, and attention mechanism fusion. The fused sequence feature is obtained by fusion of protein sequence features and target protein MSA features. The fused pairing feature is obtained by fusion of protein pairing features, target protein pairing features, protein sequence features, and target protein MSA features based on protein structure and target protein structure. Through feature fusion, information of different dimensions such as sequence and structure can be combined to improve the accuracy of fused features.

[0136] In this embodiment, a method for performing feature fusion of various features in protein sequence features and target protein features is provided, and the fusion sequence features are used to indicate the sequence information of the protein framework and the target protein sequence. The fusion pairing feature is used to indicate the information of the respective amino acid residues under the intrachain or interchain interaction between any two amino acid residues in the same peptide chain in the protein and any two amino acid residues in the same peptide chain in the target protein, and also includes the structural information and sequence information corresponding to the protein framework and the fusion protein sequence. Thereby, the interaction between the protein and the target protein can be fully expressed, which is conducive to improving the accuracy of protein design.

[0137] In some embodiments, step 362 may be optionally implemented as step 3621 and step 3622:

[0138] Step 3621, performing a linear transformation on the protein sequence features to obtain the transformed protein sequence features; and performing a linear transformation on the first row of target protein MSA features in the target protein MSA features to obtain the transformed first row of target protein MSA features.

[0139] Linear transformation can be used for at least one of feature extraction, dimensionality reduction and data conversion. The transformed protein sequence feature is obtained by performing a linear transformation on the protein sequence. The transformed first-row target protein MSA feature is obtained by performing a linear transformation on the first-row target protein MSA feature in the target protein MSA feature. In the target protein MSA data, the target protein sequence can be located in the first row.

[0140] Step 3622, based on cross attention, the transformed protein sequence features and the transformed first-row target protein MSA features are fused to obtain fused sequence features.

[0141] The Cross Attention mechanism can focus on the interaction between the transformed protein sequence features and the transformed first-line target protein MSA features. Exemplarily, the cross attention weights corresponding to the transformed protein sequence features and the transformed first-line target protein MSA features are calculated based on the cross attention, and the transformed protein sequence features and the transformed first-line target protein MSA features are weighted summed using the cross attention weights corresponding to the transformed protein sequence features and the transformed first-line target protein MSA features to obtain the fused sequence features.

[0142] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0143]

[0144] Among them, Linear is a linear transformation, It is a protein sequence feature, specifically an antibody sequence feature, It is the first-line target protein MSA feature, specifically the first-line antigen MSA feature, CrossAttention is cross attention, It is a fusion sequence feature.

[0145] In one example, the target protein is an antigen and the protein is an antibody. For cross attention, there is the following code representation:

[0146]

[0147] 1.

[0148] 2.

[0149] 3.

[0150] 4.

[0151] 5.

[0152] Among them, Linear is a linear transformation, It is a protein sequence feature, specifically an antibody sequence feature, and concat is splicing. It is the first-line target protein MSA feature, specifically the first-line antigen MSA feature, and MultiHeadAttention is multi-head attention.

[0153] In this embodiment, a method for obtaining fusion sequence features is provided, which can improve the accuracy of fusion sequence feature expression and facilitate subsequent protein sequence recovery.

[0154] In some embodiments, step 364 may be optionally implemented as step 420, step 440, and step 460:

[0155] Step 420, based on the carbon atom coordinates of each amino acid residue in the protein structure, determine the protein coordinates corresponding to each amino acid residue in the protein structure; and, based on the carbon atom coordinates of each amino acid residue in the target protein structure, determine the target protein coordinates corresponding to each amino acid residue in the target protein structure.

[0156] The carbon (C) atoms of the amino acid residues include: α carbon atoms and β carbon atoms. In this embodiment, the β carbon atom coordinates of each amino acid residue in the protein are determined as the protein coordinates corresponding to the amino acid residue, and the β carbon atom coordinates of each amino acid residue in the target protein are determined as the target protein coordinates corresponding to the amino acid residue. Both the protein coordinates and the target protein coordinates can be three-dimensional coordinates.

[0157] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0158]

[0159] Among them, GetCbAtom is to obtain the coordinates of the β carbon atom. On the left side of the equation, is the protein coordinate corresponding to the ith amino acid residue in the protein, specifically the antibody coordinate corresponding to the ith amino acid residue in the antibody, is the target protein coordinate corresponding to the i-th amino acid residue in the target protein, specifically, the antigen coordinate corresponding to the i-th amino acid residue in the antigen.

[0160] Step 440, calculating the protein pairing distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the protein structure; and calculating the target protein pairing distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the target protein structure.

[0161] In some embodiments, the distance includes a Euclidean distance, which can be calculated based on the L2 norm. The protein pairing distance can be obtained by first calculating the protein pairing Euclidean distance, and then discretizing and one-hot encoding the protein pairing Euclidean distance. The target protein pairing distance can be obtained by first calculating the target protein pairing Euclidean distance, and then discretizing and one-hot encoding the target protein pairing Euclidean distance.

[0162] Optionally, the protein pairing Euclidean distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the protein structure is calculated, the protein pairing Euclidean distance is discretized to obtain a discretized protein pairing Euclidean distance, and the discretized protein pairing Euclidean distance is one-hot encoded to obtain the protein pairing distance. Also, the target protein pairing Euclidean distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the target protein structure is calculated, the target protein pairing Euclidean distance is discretized to obtain a discretized target protein pairing Euclidean distance, and the discretized target protein pairing Euclidean distance is one-hot encoded to obtain the target protein pairing distance.

[0163] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0164]

[0165] Among them, ||·||2 is the Euclidean distance, onehot is the unique hot encoding, and v represents discretization. is the protein coordinate corresponding to the ith amino acid residue in the protein, specifically the antibody coordinate corresponding to the ith amino acid residue in the antibody, is the protein coordinate corresponding to the j-th amino acid residue in the protein, specifically the antibody coordinate corresponding to the j-th amino acid residue in the antibody, is the target protein coordinate corresponding to the ith amino acid residue in the target protein, specifically, the antigen coordinate corresponding to the ith amino acid residue in the antigen, is the target protein coordinate corresponding to the j-th amino acid residue in the target protein, specifically, the antigen coordinate corresponding to the j-th amino acid residue in the antigen. is the protein pairwise Euclidean distance, specifically the antibody pairwise Euclidean distance, is the protein pairing distance, is the target protein pairing Euclidean distance, specifically the antigen pairing Euclidean distance, is the target protein pairing distance.

[0166] Step 460, based on the protein pairing distance and the target protein pairing distance, the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature are fused to obtain a fusion pairing feature.

[0167] In this embodiment, a method for obtaining fusion pairing features is provided, which can improve the accuracy of fusion pairing features and fully express the interaction between different amino acid residues. Among them, discretization and one-hot encoding can also reduce time complexity and improve data processing efficiency.

[0168] In some embodiments, the fusion pairing feature may be a matrix. In this embodiment, the fusion pairing feature matrix is ​​divided into four matrix blocks. After the four matrix blocks are calculated respectively, the four matrix blocks are concatenated to obtain the fusion pairing feature. Step 460 is specifically implemented as step 462, step 464 and step 466:

[0169] Step 462, performing mapping based on the protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature; and performing mapping based on the target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature.

[0170] Feature mapping is used to map features from a high-dimensional feature space to a low-dimensional feature space. The first part of the fusion pairing feature is obtained by performing mapping based on the protein pairing feature and the protein pairing distance, and is the top left matrix block (top left matrix) of the matrix corresponding to the fusion pairing feature. Specifically, step 462 performs mapping based on the protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature, which is implemented as steps 462A and 462B: step 462A, performs a linear transformation on the protein pairing feature to obtain the transformed protein pairing feature; and, performs a linear transformation on the protein pairing distance to obtain the transformed protein pairing distance; step 462B, adds the transformed protein pairing feature and the transformed protein pairing distance to obtain the first part of the fusion pairing feature.

[0171] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0172]

[0173] in, It is the first part of the fusion pairing feature, Linear is a linear transformation, is a protein pairing feature, specifically an antibody pairing feature, is the protein pairing distance, specifically the antibody pairing distance. The first part of the fusion pairing feature corresponds to the matrix block size L ab ×L ab, L ab is the protein structure x ab The sequence length (SequenceLength), specifically the sequence length of the antigen structure.

[0174] The second part of the fusion pairing feature is obtained by mapping the target protein pairing feature and the target protein pairing distance, and is the bottom right matrix block of the matrix corresponding to the fusion pairing feature. Specifically, step 462 performs mapping based on the target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature, which is implemented as step 462C and step 462D: step 462C, performs a linear transformation on the target protein pairing feature to obtain the transformed target protein pairing feature; and, performs a linear transformation on the target protein pairing distance to obtain the transformed target protein pairing distance; step 462D, adds the transformed target protein pairing feature and the transformed target protein pairing distance to obtain the second part of the fusion pairing feature.

[0175] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0176]

[0177] in, The second part is the fusion pairing feature. is the target protein pairing distance, specifically the antigen pairing distance, is the target protein pairing distance, specifically the antigen pairing distance. The second part of the matrix block size corresponding to the fusion pairing feature is L ab ×L ab , L ab is the protein structure x ab The sequence length (Sequence Length), specifically the sequence length of the antigen structure.

[0178] Step 464, calculating the outer product average based on the protein sequence feature and the target protein MSA feature to obtain the third part of the fusion pairing feature; and determining the transposition of the third part of the fusion pairing feature as the fourth part of the fusion pairing feature.

[0179] The outer product mean (OuterProductMean) is used to update the information in the protein sequence features and the target protein MSA features to the pairing features, so that the protein design model can better understand and predict the protein structure. The calculation of the outer product mean usually includes: outer product calculation, mean calculation, and linear transformation. The third part of the fusion pairing feature is obtained by calculating the outer product mean based on the protein sequence features and the target protein MSA features. It is the top right matrix of the matrix corresponding to the fusion pairing feature.

[0180] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0181]

[0182] in, The third part is the fusion pairing feature. It is a protein sequence feature, specifically an antibody sequence feature, It is the target protein MSA feature, specifically the antigen MSA feature, and OuterProductMean is the outer product mean.

[0183] Specifically, step 464 is implemented as step 464A and step 464B: step 464A, sequentially performing layer normalization processing and linear transformation on the protein sequence features to obtain the transformed normalized protein sequence features; and, sequentially performing layer normalization processing and linear transformation on the target protein MSA features to obtain the transformed normalized target protein MSA features; step 464B, calculating the outer product average of each element in the transformed normalized protein sequence features and each element in the transformed normalized target protein MSA features, and, performing a linear transformation on the outer product average to obtain the third part fusion pairing features.

[0184] In one example, the target protein is an antigen and the protein is an antibody. For calculating the outer product average, there is the following code representation:

[0185]

[0186] Input:number of dimensions in hidden embeddings c

[0187] 1.

[0188] 2.

[0189] 3.

[0190] 4.

[0191] 5.

[0192] 6.z ij =Linear(o ij )

[0193] 7.return{z ij}

[0194] in, It is a protein sequence feature, specifically an antigen sequence feature, is the target protein MSA feature, specifically the antigen MSA feature, LayerNorm is layer normalization, Linear is linear transformation, a si 、b si is the transformed normalized protein sequence feature, the transformed normalized target protein MSA feature, o ij is the outer product average, z ij It is the third part that integrates the pairing features.

[0195] The fourth part of the fusion pairing feature is determined based on the symmetric property, and is the bottom left matrix block of the matrix corresponding to the fusion pairing feature. Specifically, the transposition of the third part of the fusion pairing feature is determined as the fourth part of the fusion pairing feature.

[0196] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0197]

[0198] in, The third part is the fusion pairing feature. It is the fourth part and paired feature, and T represents transposition.

[0199] Step 466, concatenate the first part of the fusion pairing features, the second part of the fusion pairing features, the third part of the fusion pairing features and the fourth part of the fusion pairing features to obtain a fusion pairing feature.

[0200] In this embodiment, specific implementation steps for obtaining fusion pairing features are provided, the accuracy of fusion pairing features is improved, and full expression of the interaction between different amino acid residues is achieved.

[0201] In some embodiments, in addition to protein features and target protein features, additional features: inter-chain features or PPI features can be input into the protein design model to provide inter-chain connection information and interaction information between the target protein and different peptide chains of the protein to further update the fusion pairing features and fusion sequence features. Optionally, after step 360, the method further includes steps 372 and 374:

[0202] Step 372, in the case where there are interchain features or interaction features between the protein and the target protein, based on the interchain features or interaction features, as well as the protein structure and the target protein structure, obtain updated sequence features and updated pairing features.

[0203] Step 374: update the fused sequence feature based on the updated sequence feature to obtain an updated fused sequence feature; and update the fused pairing feature based on the updated pairing feature to obtain an updated fused pairing feature.

[0204] The feature updating method includes: summation and weighted summation. Specifically, the update sequence feature is linearly transformed to obtain the transformed update sequence feature, and the transformed update sequence feature is added to the fusion sequence feature to obtain the updated fusion sequence feature. Also, the update pairing feature is linearly transformed to obtain the transformed update pairing feature, and the transformed update pairing feature is added to the fusion pairing feature to obtain the updated fusion pairing feature.

[0205] In one example, the target protein is an antigen and the protein is an antibody. The following codes exist:

[0206]

[0207] in, is the update sequence feature, is to update the pairing characteristics, It is the protein structure, specifically the antibody structure, It is the target protein structure, specifically the antigen structure, is the interaction characteristic, is the inter-chain feature, Linear is the linear transformation, is the fusion sequence feature, It is a fusion pairing feature.

[0208] In some embodiments, the inter-chain features and the interaction features can be converted, and one feature can be used to determine the other feature. The method of updating the fusion sequence features and fusion pairing features using the inter-chain features or interaction features, protein structure and target protein structure is also called inter-chain encoding (Inter-Chain Feature Embedding, IcfEmbedding). Optionally, step 372 is specifically implemented as steps 3721 and 3722:

[0209] Step 3721, when there are inter-chain features, determine the interaction features based on the inter-chain features and the inter-chain connection relationship between the protein structure and the target protein structure; or, when there are interaction features, determine the inter-chain features based on the interaction features and the inter-chain pseudo-connection relationship between the protein structure and the target protein structure.

[0210] In some embodiments, in the presence of inter-chain features, the protein structure and the target protein structure have an inter-chain connection relationship (inter-chain contact), and the inter-chain features and the inter-chain connection relationship are used to generate interaction features (Generate PPI using inter-chain contact). Or, in the presence of interaction features, the protein structure and the target protein structure may have an inter-chain connection relationship, called an inter-chain pseudo-connection relationship (pseudocontact), and the interaction features and the inter-chain pseudo-connection relationship are used to generate inter-chain features (Generate pseudocontact using PPI).

[0211] Step 3722, perform a linear transformation on the interaction features to obtain updated sequence features; and perform a linear transformation on the inter-chain features to update the pairing features.

[0212] In one example, the target protein is an antigen and the protein is an antibody. For the inter-chain encoding method, there are the following codes:

[0213]

[0214] 1.

[0215] 2.

[0216] 3.else

[0217] 4.

[0218] 5.end if

[0219] 6.

[0220] 7.

[0221] 8.else

[0222] 9.

[0223] 10.end if

[0224] 11.

[0225] 12.

[0226] 13.

[0227] 14.

[0228] 15.

[0229] 16.

[0230] in, is the update sequence feature, is to update the pairing characteristics, It is the protein structure, specifically the antibody structure, It is the target protein structure, specifically the antigen structure, is the interaction characteristic, is the inter-chain feature, Linear is the linear transformation, is the fusion sequence feature, It is a fusion pairing feature.

[0231] In this embodiment, when there are additional features, these additional features can also be converted into updated sequence features and updated pairing features that are similar in format to the fusion sequence features and fusion pairing features, so as to update the fusion sequence features and fusion pairing features and improve the accuracy of the fusion sequence features and fusion pairing features.

[0232] In some embodiments, combined with the above embodiments, the target protein is an antigen (Antigen, Ag) and the protein is an antibody (Antibody, Ab) as an example. The whole process of feature fusion is represented by the following code:

[0233]

[0234] Input:number of dimensions in hidden embeddings c

[0235] 1.

[0236] 2.

[0237] 3.

[0238] 4.

[0239] 5. Calculate the top left matrix block by adding the projections ofthe one-hot embedding of distance and the pairwise representation. / / Calculate the top left matrix block by mapping and adding the antibody pairing distance and antibody pairing features.

[0240] 6. The block size is L ab ×L ab ,where L ab is the sequence length of x ab .

[0241] 7.

[0242] 8. Calculate the bottom right matrix block by adding the projectionsof the one-hot embedding of distance and the pairwise representation. / / Calculate the bottom right matrix block by mapping and adding the antigen pairing distance and antigen pairing features.

[0243] 9.

[0244] 10. Calculate the top right matrix block by using OuterProductMeanbetween the single representation and / / The upper right matrix block is calculated using the outer product average between the individual feature representations of the antibody sequence features and the antigen MSA sequence features.

[0245] 11.

[0246] 12. Assign the bottom left matrix block by using symmetric propertyof / / Use the symmetry property to specify the lower left corner of the matrix block.

[0247] 13.

[0248] 14.

[0249] 15.

[0250] 16.

[0251] 17. If additional inter-chain feature provided,update viaresidual connection. / / Updated by the connection between amino acid residues when additional interchain features are provided as well as

[0252] 18.

[0253] 19.

[0254] 20.

[0255] twenty one.

[0256] 22.end if

[0257] twenty three.

[0258] In other embodiments, in addition to interchain features and interaction features, additional features may also be: epitope features and contact features. Epitope features and contact features are input into the protein design model to provide information on the key positions of the mutual binding of the target protein and the protein, and information on the spatial contact relationship. An epitope is a portion of the target protein that can bind to the protein, usually located on the surface of the target protein, and is a key position for protein recognition and binding to the target protein. Epitope features can be used to indicate the region where the epitope of the target protein is located and the amino acid sequence corresponding to the region. Contact features can be used to indicate the spatial contact relationship between different amino acid residues in a protein, and the spatial contact relationship can be determined by calculating the distance between two amino acid residues.

[0259] In some embodiments, the protein framework is a plurality of protein frameworks. In this embodiment, a multi-stage iterative prediction and screening method is set up, by performing protein sequence recovery and protein structure prediction on a plurality of protein frameworks respectively, performing screening of predicted protein sequences according to confidence scores, and reusing the screened predicted protein sequences, continuing to generate new protein frameworks and performing iterative predictions, and finally, after the iteration is completed, the predicted protein sequence and its predicted protein structure with the highest confidence score are determined as the final predicted protein sequence and its predicted protein structure. Exemplarily, after step 380, the method further includes steps 392 and 394:

[0260] Step 392, perform protein sequence recovery and protein structure prediction on the protein framework of the i-th stage in sequence, and screen out candidate protein sequences from the predicted protein sequences of the i-th stage based on the confidence scores of the predicted protein sequences and predicted protein structures respectively corresponding to the protein framework of the i-th stage; i is greater than 0.

[0261] Step 394, mask the amino acid sequence corresponding to the complementary determining region in the candidate protein sequence of the i-th stage to obtain the protein framework of the i+1-th stage; update i to i+1, and repeat the steps of performing protein sequence recovery and protein structure prediction on the protein framework of the i-th stage in sequence until the set number of iterations is reached, and the predicted protein sequence and predicted protein structure with the highest confidence score are obtained.

[0262] Since the heavy chain and light chain of a protein each have three complementary determining regions: CDR1, CDR2, and CDR3. The number of iterations in this embodiment can also be set to 3, that is, i is greater than 0 and less than or equal to 3. The protein frameworks in the protein framework library are obtained by masking the amino acid sequences corresponding to the three complementary determining regions of the heavy chain and the light chain. In subsequent iterations, the protein framework is obtained by masking the amino acid sequence corresponding to at least one of the three complementary determining regions in the candidate protein sequence.

[0263] When i=1, the protein framework of the first stage is: at least a part or all of the protein frameworks in the protein framework library are obtained by masking the amino acid sequences corresponding to CDR1, CDR2 and CDR3 of the protein sequence. Protein sequence recovery and protein structure prediction are performed on the protein framework of the first stage in turn to obtain the predicted protein sequence, predicted protein structure, and confidence score corresponding to the protein framework of the first stage. Next, based on the confidence scores of the predicted protein sequence and predicted protein structure corresponding to the protein framework of the first stage, candidate protein sequences are screened out from the predicted protein sequence of the first stage. Optionally, the candidate protein sequence can be a predicted protein sequence with a confidence score greater than a threshold, or a predicted protein sequence with a confidence score ranked in the top N, where N is greater than 0.

[0264] Next, when i=2, masking is performed on the amino acid sequences corresponding to CDR2 and CDR3 of the predicted protein sequence of the first stage to obtain the protein framework of the second stage. Protein sequence recovery and protein structure prediction are performed on the protein framework of the second stage in sequence to obtain the predicted protein sequence, predicted protein structure, and confidence score corresponding to the protein framework of the second stage. Next, based on the confidence scores of the predicted protein sequence and predicted protein structure corresponding to the protein framework of the second stage, candidate protein sequences are screened out from the predicted protein sequence of the second stage. Optionally, the candidate protein sequence can be a predicted protein sequence with a confidence score greater than a threshold, or a predicted protein sequence with a confidence score ranked in the top N, where N is greater than 0.

[0265] Finally, when i=3, the amino acid sequence corresponding to the CDR3 of the predicted protein sequence in the second stage is masked to obtain the protein framework in the third stage. Protein sequence recovery and protein structure prediction are performed on the protein framework in the third stage in turn to obtain the predicted protein sequence, predicted protein structure, and confidence score corresponding to the protein framework in the third stage. If the current iteration has reached the set number of iterations, the predicted protein sequence and predicted protein structure with the highest confidence score are selected.

[0266] In some embodiments, after obtaining the predicted protein sequence and predicted protein structure with the highest confidence score, the method may optionally include: evaluating the performance and stability of the predicted protein sequence and predicted protein structure by molecular dynamics simulation (MD).

[0267] In this embodiment, multiple iterations and cycles of designing protein sequences and protein structures and performing screening on these protein structures and protein sequences are provided. Multiple iterations can improve the accuracy of the final predicted protein structure and predicted protein sequence, thereby improving the accuracy of protein design.

[0268] In some embodiments, a protein framework library is pre-constructed, and the protein framework library includes a plurality of protein frameworks. Prior to step 320, the method further includes steps 312, 314, and 316:

[0269] Step 312: Encode the protein sequence in the protein sequence database to obtain an encoded protein sequence.

[0270] Protein sequence databases refer to some public databases, such as SWISS-PROT, UniPot, and SCOP databases. Taking the case of antibodies as an example, the antibody sequence database can be: Observed Antibody Space database (OAS). The encoding method can be implemented by numbering, for example, numbering each amino acid in the protein sequence. The encoded protein sequence can indicate the position, structure, and type of each amino acid.

[0271] Step 314, compare the encoded protein sequence with the standard database to obtain the variable region-joint region genes and frequencies corresponding to the encoded protein sequence.

[0272] The standard database is a database dedicated to immunology and genetics research, which provides a large amount of gene and protein sequence information, as well as related annotations and function predictions. The standard database includes the IMGT database. Taking the protein antibody as an example, the antibody structure includes: variable region (V region) and joining region (J region). The variable region is used to recognize and bind to a specific type of antigen, and the joining region is used to ensure the stability of the antibody structure. The variable region-joining region gene (VJ gene) and frequency can indicate the complementary determining region in the antibody.

[0273] Step 316, determining the protein framework based on the encoded protein sequence with a frequency greater than a threshold.

[0274] In some embodiments, the region where the frequency of the variable region-joining region gene is greater than a threshold is determined as a complementary determining region. For the original protein sequence corresponding to the encoded protein sequence, the complementary determining region in the protein sequence is masked to obtain a protein framework.

[0275] In this embodiment, a method for pre-constructing protein frameworks is provided. These protein frameworks are essential for the stability and solubility of proteins and can be used for protein sequence recovery, thereby making protein design a high-throughput protein design method and improving the efficiency of protein design.

[0276] Part III: Training phase of protein design models

[0277] In some embodiments, the training phase of the protein design model is performed by a computer device, which may be Figure 1 The terminal 120 and / or the server 140 store the protein design model. Before step 320, the method further includes at least some of steps 520, 540, 560, and 580:

[0278] Step 520: Obtain a sample protein sequence, and extract sample protein features from the sample protein sequence.

[0279] The sample protein sequence is a complete protein sequence used in the training phase of the protein design model, and is used to provide complete sequence information for the learning of the protein design model. The sample protein sequence is a sequence composed of peptide chains and amino acids on the peptide chains in the sample protein. The sample protein sequence includes the amino acid sequence corresponding to the heavy chain and the amino acid sequence corresponding to the light chain. For example, the sample protein sequence is represented as "H: QIQLVO...; L: DVLMTQ...YKDDDK", "H" is the heavy chain, "L" is the light chain, and one of the letters "QILVOCARWGSDMTYK" is used to represent a type of amino acid.

[0280] The sample protein feature is a feature corresponding to the sample protein sequence. In this embodiment, the sample protein features of the sample protein are extracted from at least two dimensions: sequence and structure. Optionally, the sample protein features include: sample protein sequence features (Sequence Representation, s), sample protein pairing features (Pairwise Representation, z) and sample protein structure (x). Among them, the sample protein sequence feature is used to indicate the sequence information of the sample protein sequence. The sample protein pairing feature can also be called the sample protein paired feature, which is a feature corresponding to any two amino acid residues located in the same peptide chain, and is used to indicate the interaction between any two amino acid residues located in the same peptide chain. The sample protein structure is represented by the atomic coordinates of each amino acid residue in the sample protein, which is used to indicate the structural information of the sample protein. The atomic coordinates can be three-dimensional coordinates.

[0281] Step 540, obtaining a sample target protein sequence, and extracting a sample target protein feature of the sample target protein sequence.

[0282] The sample target protein sequence is the target protein sequence used in the training phase of the protein design model. The sample target protein sequence is a sequence consisting of a peptide chain in the sample target protein and amino acids on the peptide chain. For example, the sample target protein includes an antigen, and the sample target protein sequence is represented as "A: RVQPTE...", "A" represents the antigen, and one of the letters in "RVQPTE" represents a type of amino acid.

[0283] In some embodiments, a sample target protein sequence is obtained, a sequence database is searched based on the sample target protein sequence, at least one homologous sequence related to the sample target protein sequence is determined, the sample target protein sequence and at least one homologous sequence related to the sample target protein sequence are combined to obtain sample target protein multiple sequence alignment (MSA) data, and sample target protein features of the sample target protein MSA data are extracted. Optionally, when combining the sample target protein sequence and at least one homologous sequence related to the sample target protein sequence to obtain the sample target protein MSA data, the sample target protein sequence can be placed in the first row of the sample target protein MSA data.

[0284] Among them, the sample target protein MSA data includes: the sample target protein sequence and at least one homologous sequence related to the sample target protein sequence. Through these homologous sequences, the structural and functional related characteristics of the sample target protein retained during the evolution process can be captured. For example, the sample target protein includes an antigen, and the sample target protein sequence is represented as "A: RVQPTE...". After searching the sequence database, three homologous sequences are determined, which are represented as "RVVPSG...", "RVQPTV...", and "RVSPST...". The sample target protein sequence and the three homologous sequences are combined to obtain the sample target protein MSA data.

[0285] The sample target protein feature is the feature corresponding to the sample target protein sequence. In this embodiment, the sample target protein feature of the sample target protein sequence is extracted from at least two dimensions, sequence and structure. Optionally, the sample target protein feature includes: sample target protein MSA feature (MSA Representation, m), sample target protein pairing feature (PairwiseRepresentation, z) and sample target protein structure (x). Among them, the sample target protein MSA feature is used to indicate the sequence information of the sample target protein sequence and its homologous sequence. The sample target protein pairing feature can also be called the sample target protein paired feature, which is the feature corresponding to any two amino acid residues located in the same peptide chain, and is used to indicate the interaction between any two amino acid residues located in the same peptide chain. The sample target protein structure is represented by the atomic coordinates of each amino acid residue in the target protein, which is used to indicate the structural information of the target protein. The atomic coordinates can be three-dimensional coordinates.

[0286] Step 560, fusing the sample protein feature with the sample target protein feature to obtain a sample fusion feature; the sample fusion feature is used to characterize the sequence information and structural information under the interaction between the sample protein indicated by the sample protein sequence and the sample target protein indicated by the sample target protein sequence.

[0287] The sample fusion feature (Complex Representation) is obtained by performing feature fusion (Feature Fusion) on the sample protein feature and the sample target protein feature. Since there is an interaction between the sample protein and the sample target protein, the two affect each other, so the sample fusion feature is used to characterize the sequence information and structural information under the interaction between the sample protein indicated by the sample protein sequence and the sample target protein indicated by the sample target protein sequence.

[0288] In this embodiment, the sample fusion feature includes: a sample fusion sequence feature and a sample fusion pairing feature. Among them, the sample fusion sequence feature is used to indicate the sequence information under the interaction between the sample protein and the sample target protein. The sample fusion pairing feature is used to indicate the information of the respective amino acid residues under the intrachain or interchain interaction between any two amino acid residues located in the same peptide chain in the sample protein and any two amino acid residues located in the same peptide chain in the sample target protein. The sample fusion pairing feature also includes the structural information and sequence information corresponding to the sample protein sequence and the sample fusion protein sequence, which will be explained in subsequent embodiments.

[0289] Step 580, perform prediction on the sample fusion features to obtain prediction results, and adjust the model parameters of the protein design model with the training goal of reducing the difference between the prediction results and the sample results.

[0290] The prediction performed in this embodiment is a multi-task prediction, and each task includes a corresponding prediction result. The model parameters of the protein design model are adjusted with the training goal of reducing the difference between the prediction result and the sample result.

[0291] In summary, the protein design method provided in the embodiments of the present application obtains a sample protein sequence and extracts sample protein features of the sample protein sequence; obtains a sample target protein sequence and extracts sample target protein features of the sample target protein sequence; fuses the sample protein feature with the sample target protein feature to obtain a sample fusion feature; the sample fusion feature is used to characterize the sequence information and structural information under the interaction between the sample protein indicated by the sample protein sequence and the sample target protein indicated by the sample target protein sequence; performs prediction on the sample fusion feature to obtain a prediction result, and adjusts the model parameters of the protein design model with the training goal of reducing the difference between the prediction result and the sample result. Accordingly, a training method for a protein design model is provided, which can utilize sample protein sequences and sample target protein sequences, and by fusing the sample protein features of the sample protein sequences and the sample target protein features of the sample target protein sequences, the protein design model can fully learn the sample target proteins indicated by the sample target protein sequences, and the interactions between the sample proteins corresponding to the sample protein sequences, and fully learn various features, thereby improving the accuracy of the protein design model, so that the protein design model can design proteins with the ability to bind to and interact with target proteins during the application stage, thereby improving the accuracy of protein design, and ensuring the rationality of the protein structure while designing the protein sequence.

[0292] In some embodiments, the sample protein features include: sample protein sequence features, sample protein pairing features and sample protein structure, the sample target protein features include: sample target protein MSA features, sample target protein pairing features and sample target protein structure, and the sample fusion features include: sample fusion sequence features and sample fusion pairing features. Exemplarily, step 560 is specifically implemented as steps 562 and 564:

[0293] Step 562, fusion of the sample protein sequence feature and the sample target protein MSA feature to obtain the sample fusion sequence feature.

[0294] Step 564, based on the sample protein structure and the sample target protein structure, the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature are fused to obtain the sample fusion pairing feature.

[0295] The feature fusion methods include: at least one of addition, multiplication, concatenation, linear combination, weighted summation, and attention mechanism fusion. The sample fusion sequence feature is obtained by fusion of the sample protein sequence feature and the sample target protein MSA feature. The sample fusion pairing feature is obtained by fusion of the sample protein structure and the sample target protein structure, and the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature, and the sample target protein MSA feature. Through feature fusion, information of different dimensions such as sequence and structure can be combined to improve the accuracy of the sample fusion feature.

[0296] In this embodiment, a method for performing feature fusion of various features in the sample protein sequence features and the sample target protein features is provided, and the sample fusion sequence features are used to indicate the sequence information of the sample protein sequence and the sample target protein sequence. The sample fusion pairing features are used to indicate the information of the respective amino acid residues under the intrachain or interchain interaction between any two amino acid residues in the sample protein and any two amino acid residues in the same peptide chain in the sample target protein, and also include the structural information and sequence information corresponding to the sample protein sequence and the sample fusion protein sequence. Thereby, the interaction between the sample protein and the sample target protein can be fully expressed, which is conducive to the full learning of the protein design model.

[0297] In some embodiments, step 562 may be optionally implemented as steps 620 and 640:

[0298] Step 620, performing a linear transformation on the sample protein sequence features to obtain the transformed sample protein sequence features; and performing a linear transformation on the first row of sample target protein MSA features in the sample target protein MSA features to obtain the transformed first row of sample target protein MSA features.

[0299] Linear transformation (Linear) can be used for at least one of feature extraction, dimensionality reduction and data conversion. The transformed sample protein sequence feature is obtained by performing a linear transformation on the sample protein sequence. The transformed first-row sample target protein MSA feature is obtained by performing a linear transformation on the first-row sample target protein MSA feature in the sample target protein MSA feature. In the sample target protein MSA data, the sample target protein sequence can be located in the first row.

[0300] Step 640, based on cross attention, the transformed sample protein sequence features and the transformed first-row sample target protein MSA features are fused to obtain sample fusion sequence features.

[0301] The Cross Attention mechanism can focus on the interaction between the transformed sample protein sequence features and the transformed first-line sample target protein MSA features. Exemplarily, the cross attention weights corresponding to the transformed sample protein sequence features and the transformed first-line sample target protein MSA features are calculated based on the cross attention, and the transformed sample protein sequence features and the transformed first-line sample target protein MSA features are weighted summed using the cross attention weights corresponding to the transformed sample protein sequence features and the transformed first-line sample target protein MSA features to obtain the sample fusion sequence features.

[0302] In this embodiment, a method for obtaining sample fusion sequence features is provided, which can improve the accuracy of sample fusion sequence feature expression and is conducive to protein sequence recovery in the application stage of the protein design model.

[0303] In some embodiments, step 564 may be optionally implemented as step 620, step 640, and step 660:

[0304] Step 620, based on the carbon atom coordinates of each amino acid residue in the sample protein structure, determine the protein coordinates corresponding to each amino acid residue in the sample protein structure; and, based on the carbon atom coordinates of each amino acid residue in the sample target protein structure, determine the target protein coordinates corresponding to each amino acid residue in the sample target protein structure.

[0305] The carbon (C) atoms of the amino acid residues include: α carbon atoms and β carbon atoms. In this embodiment, the β carbon atom coordinates of each amino acid residue in the sample protein are determined as the protein coordinates corresponding to the amino acid residue, and the β carbon atom coordinates of each amino acid residue in the sample target protein are determined as the target protein coordinates corresponding to the amino acid residue. Both the protein coordinates and the target protein coordinates can be three-dimensional coordinates.

[0306] Step 640, calculating the protein pairing distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample protein structure; and calculating the target protein pairing distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample target protein structure.

[0307] In some embodiments, the distance includes a Euclidean distance, which can be calculated based on the L2 norm. The protein pairing distance can be obtained by first calculating the protein pairing Euclidean distance, and then discretizing and one-hot encoding the protein pairing Euclidean distance. The target protein pairing distance can be obtained by first calculating the target protein pairing Euclidean distance, and then discretizing and one-hot encoding the target protein pairing Euclidean distance.

[0308] Optionally, the protein pairing Euclidean distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample protein structure is calculated, the protein pairing Euclidean distance is discretized to obtain a discretized protein pairing Euclidean distance, and the discretized protein pairing Euclidean distance is one-hot encoded to obtain the protein pairing distance. Also, the target protein pairing Euclidean distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample target protein structure is calculated, the target protein pairing Euclidean distance is discretized to obtain a discretized target protein pairing Euclidean distance, and the discretized target protein pairing Euclidean distance is one-hot encoded to obtain the target protein pairing distance.

[0309] Step 660, based on the protein pairing distance and the target protein pairing distance, the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature are fused to obtain the sample fusion pairing feature.

[0310] In this embodiment, a method for obtaining sample fusion pairing features is provided, which can improve the accuracy of sample fusion pairing features and fully express the interaction between different amino acid residues. Among them, discretization and one-hot encoding can also reduce time complexity and improve data processing efficiency.

[0311] In some embodiments, the sample fusion pairing feature may be a matrix. In this embodiment, the sample fusion pairing feature matrix is ​​divided into four matrix blocks. After calculating the four matrix blocks respectively, the four matrix blocks are spliced ​​to obtain the sample fusion pairing feature. Step 660 is specifically implemented as step 662, step 664 and step 666:

[0312] Step 662, performing mapping based on the sample protein pairing features and the protein pairing distance to calculate the first part of the fusion pairing features; and performing mapping based on the sample target protein pairing features and the target protein pairing distance to calculate the second part of the fusion pairing features.

[0313] Feature mapping is used to map features from a high-dimensional feature space to a low-dimensional feature space. The first part of the fusion pairing feature is obtained by mapping based on the sample protein pairing feature and the protein pairing distance, and is the top left matrix block of the matrix corresponding to the sample fusion pairing feature. Specifically, step 662 performs mapping based on the sample protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature, which is implemented as steps 662A and 662B: Step 662A, performs a linear transformation on the sample protein pairing feature to obtain the transformed sample protein pairing feature; and, performs a linear transformation on the protein pairing distance to obtain the transformed protein pairing distance; Step 662B, adds the transformed sample protein pairing feature and the transformed protein pairing distance to obtain the first part of the fusion pairing feature.

[0314] The second part of the fusion pairing feature is obtained by mapping the sample target protein pairing feature and the target protein pairing distance, and is the bottom right matrix block of the matrix corresponding to the sample fusion pairing feature. Specifically, step 662 performs mapping based on the sample target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature, which is implemented as step 662C and step 662D: step 662C, performs a linear transformation on the sample target protein pairing feature to obtain the transformed sample target protein pairing feature; and, performs a linear transformation on the target protein pairing distance to obtain the transformed target protein pairing distance; step 662D, adds the transformed sample target protein pairing feature and the transformed target protein pairing distance to obtain the second part of the fusion pairing feature.

[0315] Step 664, calculating the outer product average based on the sample protein sequence feature and the sample target protein MSA feature to obtain the third part of the fusion pairing feature; and determining the transposition of the third part of the fusion pairing feature as the fourth part of the fusion pairing feature.

[0316] The outer product mean (OuterProductMean) is used to update the information in the sample protein sequence features and the sample target protein MSA features to the pairing features, so that the protein design model can better understand and predict the protein structure. The calculation of the outer product mean usually includes: outer product calculation, mean value calculation, and linear transformation. The third part, the fusion pairing feature, is obtained by calculating the outer product mean based on the sample protein sequence features and the sample target protein MSA features, and is the top right matrix block of the matrix corresponding to the sample fusion pairing feature.

[0317] Specifically, step 664 is implemented as step 664A and step 664B: step 664A, performing layer normalization processing and linear transformation on the sample protein sequence features in sequence to obtain the transformed normalized protein sequence features; and, performing layer normalization processing and linear transformation on the sample target protein MSA features in sequence to obtain the transformed normalized target protein MSA features; step 664B, calculating the outer product average of each element in the transformed normalized protein sequence features and each element in the transformed normalized target protein MSA features, and performing a linear transformation on the outer product average to obtain the third part fusion pairing features.

[0318] The fourth part of the fusion pairing feature is determined based on the symmetric property, and is the bottom left matrix block of the matrix corresponding to the sample fusion pairing feature. Specifically, the transposition of the third part of the fusion pairing feature is determined as the fourth part of the fusion pairing feature.

[0319] Step 666, concatenate the first part of the fusion pairing features, the second part of the fusion pairing features, the third part of the fusion pairing features and the fourth part of the fusion pairing features to obtain the sample fusion pairing features.

[0320] In this embodiment, specific implementation steps for obtaining sample fusion pairing features are provided, which improves the accuracy of sample fusion pairing features and achieves full expression of the interaction between different amino acid residues.

[0321] In some embodiments, in addition to the sample protein features and the sample target protein features, additional features: sample inter-chain features (Inter-chain Feature) or sample interaction features (PPI Feature) can also be input into the protein design model to provide inter-chain connection information and interaction information between the sample target protein and different peptide chains of the sample protein, so as to further update the sample fusion pairing features and the sample fusion sequence features. Optionally, after step 560, the method further includes steps 572 and 574:

[0322] Step 572, in the presence of a sample interchain feature or a sample interaction feature between the sample protein and the sample target protein, based on the sample interchain feature or the sample interaction feature, as well as the sample protein structure and the sample target protein structure, obtain a sample updated sequence feature and a sample updated pairing feature.

[0323] Step 574: update the sample fusion sequence feature based on the sample update sequence feature to obtain an updated sample fusion sequence feature; and update the sample fusion pairing feature based on the sample update pairing feature to obtain an updated sample fusion pairing feature.

[0324] The feature updating method includes: summation and weighted summation. Specifically, the sample update sequence feature is linearly transformed to obtain the transformed sample update sequence feature, and the transformed sample update sequence feature is added to the sample fusion sequence feature to obtain the updated sample fusion sequence feature. Also, the sample update pairing feature is linearly transformed to obtain the transformed sample update pairing feature, and the transformed sample update pairing feature is added to the sample fusion pairing feature to obtain the updated sample fusion pairing feature.

[0325] In some embodiments, the sample inter-chain features and the sample interaction features can be converted, and one feature can be used to determine the other feature. The method of updating the sample fusion sequence features and the sample fusion pairing features using the sample inter-chain features or the sample interaction features, the sample protein structure and the sample target protein structure is also called the inter-chain encoding method. Optionally, step 572 is implemented as steps 5721 and 5722:

[0326] Step 5721, when there are sample inter-chain features, determine the sample interaction features based on the sample inter-chain features and the inter-chain connection relationship; or, when there are sample interaction features, determine the sample inter-chain features based on the sample interaction features and the inter-chain pseudo-connection relationship.

[0327] In some embodiments, when there are sample inter-chain features, the sample protein structure and the sample target protein structure have an inter-chain connection relationship (inter-chain contact), and the sample inter-chain features and the inter-chain connection relationship are used to generate sample interaction features (Generate PPI using inter-chain contact). Or, when there are sample interaction features, the sample protein structure and the sample target protein structure may have an inter-chain connection relationship, called an inter-chain pseudo-connection relationship (pseudo contact), and the sample interaction features and the inter-chain pseudo-connection relationship are used to generate sample inter-chain features (Generate pseudo contact using PPI).

[0328] Step 5722, perform linear transformation on the sample interaction features to obtain sample update sequence features; and perform linear transformation on the sample inter-chain features to obtain sample update pairing features.

[0329] In this embodiment, if there are additional features, these additional features can also be converted into sample update sequence features and sample update pairing features that are similar in format to the sample fusion sequence features and sample fusion pairing features, so as to update the sample fusion sequence features and sample fusion pairing features and improve the accuracy of the sample fusion sequence features and sample fusion pairing features.

[0330] In other embodiments, in addition to the sample interchain features and sample interaction features, additional features may also be: sample epitope features, sample contact features. The sample epitope features and sample contact features are input into the protein design model to provide information on the key positions of the mutual binding between the sample target protein and the sample protein, and information on the spatial contact relationship. An epitope is a portion of the target protein that can bind to a protein, usually located on the surface of the target protein, and is a key position for protein recognition and binding to the target protein. The sample epitope feature can be used to indicate the region where the sample target protein epitope is located and the amino acid sequence corresponding to the region. The sample contact feature can be used to indicate the spatial contact relationship between different amino acid residues in the sample protein, and the spatial contact relationship can be determined by calculating the distance between two amino acid residues.

[0331] It should also be noted that for the code representation of the feature fusion part in the training stage of the protein design model, reference can be made to the code representation of the feature fusion part in the application stage of the protein design model in the aforementioned embodiment, and the code representation in the training stage of the protein design model will not be repeated.

[0332] In some embodiments, the protein design model is a multi-task model trained by multi-task learning. These tasks include: complex structure prediction task, protein sequence recovery task, protein structure prediction task. Among them, the protein sequence recovery task and the protein structure prediction task can be performed as collaborative tasks simultaneously. During the training phase of the protein design model, the prediction is a multi-task prediction. The prediction results include: predicting complex structure, predicting protein sequence and predicting protein structure, and the sample results include: sample complex structure, sample protein sequence and sample protein structure. Exemplarily, step 580 is specifically implemented as steps 582, 584, 586 and 588:

[0333] Step 582, perform complex structure prediction on the sample fusion features to obtain a predicted complex structure.

[0334] Step 584 , masking is performed on the amino acid sequence corresponding to the complementarity determining region in the sample protein sequence to obtain a masked sample protein sequence.

[0335] Step 586, performing protein sequence recovery on the sample fusion features to recover the predicted protein sequence corresponding to the masked sample protein sequence, and performing protein structure prediction on the predicted protein sequence to obtain a predicted protein structure.

[0336] Step 588, adjusting the model parameters of the protein design model with the training goal of reducing the difference between the predicted complex structure and the sample complex structure, the difference between the predicted protein sequence and the sample protein sequence, and the difference between the predicted protein structure and the sample protein structure.

[0337] Specifically, for the complex structure prediction task: perform complex structure prediction on the sample fusion feature to obtain a predicted complex structure. For the protein sequence recovery task and the protein structure prediction task: perform masking on the amino acid sequence corresponding to the complementary determining region in the sample protein sequence to obtain a masked sample protein sequence, wherein a mask is performed on the amino acid sequence corresponding to at least one complementary determining region in the sample protein sequence. Perform protein sequence recovery on the sample fusion feature to restore the predicted protein sequence corresponding to the masked sample protein sequence, and perform protein structure prediction on the predicted protein sequence to obtain a predicted protein structure. Finally, the model parameters of the protein design model are adjusted with the training objectives of reducing the difference between the predicted complex structure and the sample complex structure, the difference between the predicted protein sequence and the sample protein sequence, and the difference between the predicted protein structure and the sample protein structure.

[0338] In some embodiments, for multi-task learning of protein design models, each task can be performed separately or simultaneously. When each task is performed separately, the training goal is to reduce the loss function corresponding to the task, and the model parameters of the protein design model are adjusted. When each task is performed simultaneously, the training goal is to reduce the total loss function corresponding to multiple tasks, and the model parameters of the protein design model are adjusted.

[0339] In this embodiment, a multi-task learning method for the protein design model is provided, which utilizes the correlation between different tasks to enable the protein design model to learn multiple related tasks at the same time, improve the performance of the protein design model on each task, and enhance the learning and generalization capabilities of the protein design model.

[0340] In some embodiments, step 588 is specifically implemented as step 5881, step 5882, and step 5883:

[0341] Step 5881, based on the predicted complex structure and the sample complex structure, determine the complex loss function, the complex loss function is obtained by weighted summing at least one of the following losses: geometric loss between amino acid residues for reducing the distance and angle between amino acid residues, frame alignment point error loss for preventing local structure destruction, interface frame alignment point error loss for constraining inter-chain amino acid residue pairs, coordinate root mean square difference loss for aligning coordinates, confidence loss for estimating confidence scores, and structural violation loss for handling structural violations.

[0342] Inter-Residue Geometric Loss is used to reduce the distance and angle between amino acid residues. It is calculated based on the distance and angle between the sample α carbon atom and the sample β carbon atom of each amino acid residue in the sample complex structure and the predicted α carbon atom and the predicted β carbon atom of each amino acid residue in the predicted complex structure. Frame Aligned Point Error is used to prevent local structural damage. It is obtained by projecting each atom in the predicted complex structure onto the real sample complex structure and the predicted complex structure. Interface Frame Aligned Point Error is used to constrain interchain amino acid residue pairs. Coordinate RMSD Loss is used to align coordinates. It is determined based on the coordinate root mean square difference (RMSD) between the aligned real sample complex structure and the predicted complex structure. Confidence loss is a loss function used to estimate the confidence score. The confidence score can be: ipLDDT, ipTM score. Structure Violation Loss is used to deal with structural violations, which are incorrect peptide bond lengths and angles and steric clashes between non-bonded atoms.

[0343] In one example, the composite loss function has the following representation:

[0344] L1=L geo +L fape +L ifape +L rmsd +λ conf ·L conf +λ viol ·L viol

[0345] Among them, L1 is the composite loss function, L geo is the geometric loss between amino acid residues, L fape is the frame alignment point error loss, L ifape is the interface frame alignment point error loss, L rmsd is the coordinate root mean square error loss, L conf is the confidence loss, L viol is the structural violation loss. conf is the confidence loss weight, for example, set to 0.1, λ viol is the weight of the structural violation loss, for example, set to 0.01.

[0346] Step 5882, determining a structure loss function based on the predicted protein structure and the sample protein structure, and determining a sequence loss function based on the predicted protein sequence and the sample protein sequence.

[0347] The structure loss function is used to reduce the difference between the predicted protein structure and the sample protein structure. The sequence loss function is used to reduce the difference between the predicted protein sequence and the sample protein sequence. During the training phase, the protein design model optimizes both protein structure prediction and protein sequence recovery.

[0348] In one example, the structure loss function and the sequence loss function are expressed together as follows:

[0349] L2=λ struture ·L struture +λ sequence ·L sequence

[0350] Among them, L2 is the structure sequence collaborative loss function, L struture is the structural loss function, λ struture is the weight of the structural loss function, L sequence is the sequence loss function, λ sequence is the sequence loss function weight.

[0351] In some embodiments, the structural loss function can also be obtained by weighted summing at least one of the following losses: geometric loss between amino acid residues for reducing the distance and angle between amino acid residues, frame alignment point error loss for preventing local structure destruction, interface frame alignment point error loss for constraining inter-chain amino acid residue pairs, coordinate root mean square difference loss for aligning coordinates, confidence loss for estimating confidence scores, and structural violation loss for handling structural violations.

[0352] The geometric loss between amino acid residues is used to reduce the distance and angle between amino acid residues. It is calculated based on the distance and angle between the sample α carbon atom and the sample β carbon atom of each amino acid residue in the sample protein structure and the predicted α carbon atom and the predicted β carbon atom of each amino acid residue in the predicted protein structure. The frame alignment point error loss is used to prevent local structural damage. It is obtained by projecting each atom in the predicted protein structure onto the real sample protein structure and the predicted protein structure. The interface frame alignment point error loss is used to constrain the interchain amino acid residue pairs. The coordinate root mean square difference loss is used to align coordinates. It is determined based on the coordinate root mean square difference between the aligned real sample protein structure and the predicted protein structure. The confidence loss is a loss function used to estimate the confidence score. The confidence score can be: ipLDDT, ipTM score. The structural violation loss is used to deal with structural violations. Structural violations refer to incorrect peptide bond lengths and angles and spatial conflicts between non-bonded atoms.

[0353] Step 5883, perform weighted summation on the complex loss function, the structure loss function and the sequence loss function to obtain the total loss function; adjust the model parameters of the protein design model with the reduction of the total loss function as the training goal.

[0354] The total loss function is obtained by performing a weighted summation of the complex loss function, the structure loss function, and the sequence loss function, and is used to indicate the multi-task learning of the protein design model. With the reduction of the total loss function as the training goal, the model parameters of the protein design model are adjusted. Specifically, the model parameters corresponding to the protein feature generation unit, the target protein feature generation unit, the feature fusion unit, the complex structure prediction unit, the protein sequence recovery unit, and the protein structure prediction unit of the protein design model are adjusted.

[0355] In one example, the total loss function has the following representation:

[0356] L total =λ1·L1+λ2·L2

[0357] Among them, L total is the total loss function, λ1 is the weight of the composite loss function, L1 is the composite loss function, L2 is the structure-sequence synergy loss function, and λ2 is the weight of the structure-sequence synergy loss function.

[0358] In this embodiment, the loss function of each task and the total loss function of the multi-task learning of the protein design model are provided. Taking reducing the total loss function as the training goal, the multi-task learning of the protein design model can be realized, and the rationality of the predicted protein structure and the feasibility of the biological function can be guaranteed while generating the predicted protein sequence.

[0359] As an example, taking the target protein in the aforementioned embodiments as an antigen (Antigen, Ag) and the protein as an antibody (Antibody, Ab) as an example, the protein design method is described in detail in combination with various schematic diagrams.

[0360] (I) Application scenarios

[0361] Antibodies are the main type of protein therapy. Antibody design includes antibody sequence design and antibody structure design, which can be widely used in medicine and drug development, biotechnology, industrial applications, materials science, diagnosis and treatment and other fields. In the related art, the development of therapeutic antibodies still relies on animal immunization or screening of antibody libraries to identify candidate molecules that bind to target antigens. These methods are time-consuming and labor-intensive, and may not accurately produce antibodies that interact with target antigens. By adopting the methods of the embodiments of the present application, antibody sequences and antibody structures can be accurately designed.

[0362] 2. Protein Design Model

[0363] (1) Brief description of model structure and function

[0364] Figure 6 It is a schematic block diagram of a protein design model provided by an exemplary embodiment of the present application.

[0365] When the target protein is an antigen and the protein is an antibody, the protein design model is called: antibody design model (tFold-Ag) 300. The antibody design model (tFold-Ag) 300 includes: a pre-trained language module (ESM-PPI) (Pretrained ESM-PPI Module) 11, a pre-trained sequence-based protein structure prediction module (tFold-Ab) (Pretrained tFold-Ab Module) 12 in series; a pre-trained protein structure prediction module (Pretrained ProteinStructure Prediction Module) 21, wherein the pre-trained protein structure prediction module 21 is connected in parallel with the pre-trained ESM-PPI module 11 and the pre-trained tFold-Ab module 12; a feature fusion module (Feature Fusion Module) 31 connected to the pre-trained tFold-Ab module 12 and the pre-trained protein structure prediction module 21; a complex structure prediction module (Complex Prediction Module) 41 and a sequence recovery module 51 connected to the feature fusion module 31, wherein the complex structure prediction module 41 and the sequence recovery module 51 are connected in parallel; a structure prediction module 61 connected to the sequence recovery module 51. Optionally include: a recycling module 71 connected to both ends of the pre-trained protein structure prediction module 21; a recycling module 81 connected to both ends of the complex structure prediction module 41; a recycling module 91 connected between the sequence recovery module 51 and the pre-trained ESM-PPI module 11. In some examples, it may also be optionally included: a recycling module connected to the pre-trained ESM-PPI module 11 and the pre-trained tFold-Ab module 12, Figure 6 The feature fusion module 31, the complex structure prediction module 41, the sequence recovery unit 51 and the structure prediction unit 61 can be combined into an AI-driven flexible docking module.

[0366] Among them, the pre-trained ESM-PPI module 11 and the pre-trained tFold-Ab module 12 are used to extract antibody features; the pre-trained protein structure prediction module 21 is used to extract antigen features; the feature fusion module 31 is used to fuse antigen features and antibody features; the complex structure prediction module 41 is used to perform complex structure prediction; the sequence recovery module 51 is used to perform antibody sequence recovery; the structure prediction module 61 is used to perform antibody structure prediction; the loop module 71 and the loop module 81 are used to reuse the features and structures of previous iterations; the loop module 91 is used to perform masking and multi-stage iterative loops.

[0367] In one possible implementation, the pre-trained protein structure prediction module 21 is implemented as the Evoformer module and the Structure module in the Alphafold2 model. The sequence recovery module 51 can be implemented as a multilayer perceptron (MLP). The complex structure prediction module 41 and the structure prediction module 61 can be implemented as the Evoformer module and the Structure module in the AlphaFold2 model, respectively. Or, the complex structure prediction module 41 can also be implemented using a diffusion model to sample more possible docking posture information of the antibody and the antigen. The pre-trained ESM-PPI module 11 and the pre-trained tFold-Ab module 12 are implemented as modules in the tFold-Ab model. Optionally, the pre-trained ESM-PPI module 11 is used to extract sequence information and intra-chain information of the antibody, and generate features for downstream structure prediction.

[0368] Extraction of sequence features:

[0369] h=ESM-PPI(x)

[0370] Among them, x is the input antibody sequence and h is the extracted sequence feature.

[0371] Extraction of chain information:

[0372] h ij =Attention(h i ,h j )

[0373] Among them, h i and h j Represents the sequence characteristics of the i-th amino acid residue and the j-th amino acid residue in the same peptide chain.

[0374] Optionally, the pre-trained tFold-Ab module 12 can be specifically implemented as an Evoformer-Single module, which is used to update and optimize the sequence features input from the pre-trained ESM-PPI module 11 to generate sequence features and pairing features with richer information.

[0375] Update of sequence features:

[0376] h′=Evoformer-Single(h)

[0377] Among them, h′ is the sequence feature after being updated by the Evoformer-Single module.

[0378] Updates to pairing features:

[0379] z ij =PairUpdate(z ij ,h i ,h j )

[0380] Among them, z ij is the pairing characteristic of the ith amino acid residue and the jth amino acid residue in the same peptide chain, h i and h j Represents the sequence characteristics of the i-th amino acid residue and the j-th amino acid residue in the same peptide chain.

[0381] Optionally, the pre-trained tFold-Ab module 12 can also be implemented as a structure module, which uses the invariant point attention mechanism (IPA) to map the optimized sequence features and pairing features into a predicted atomic-level three-dimensional structure, that is, to obtain a predicted antibody structure, which serves as the initial structure of the antibody so that the feature fusion module 31 can continue to perform subsequent feature fusion.

[0382] Invariant Point Attention Mechanism:

[0383]

[0384] Among them, R i is the three-dimensional coordinate of the i-th atom in the antibody, w i is the weight corresponding to the i-th atom, calculated through the attention mechanism.

[0385] Structure Mapping:

[0386] R′=IPA(h′,R)

[0387] Wherein, R' is the antibody structure.

[0388] Optionally, when there is also a loop module connected to the pre-trained ESM-PPI module 11 and the pre-trained tFold-Ab module 12, the loop module allows the reuse of sequence features and antibody structures of previous iterations to improve the quality of the determined antibody structure.

[0389] Cycle Update:

[0390] h (t+1) =Recycle(h (t) ,R (t) )

[0391] R (t+1) =StructureUpdate(h (t+1) )

[0392] Among them, h (t+1) and R (t+1) They represent the sequence features and antibody structures after the t+1th iteration respectively.

[0393] (2) Training phase

[0394] The antibody design model (tFold-Ag) 300 is a multi-task model, and these tasks include: complex structure prediction task, antibody sequence recovery task, and antibody structure prediction task. Among them, the antibody sequence recovery task and the antibody structure prediction task can be performed as collaborative tasks at the same time.

[0395] 2.1: Complex structure prediction task

[0396] Figure 7 It is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application.

[0397] The input data of the antibody design model (tFold-Ag) 300 are: complete antibody sequence (AntibodySequence), antigen sequence (AntigenSequence), and the output data are: predicted complex structure and its confidence score. Optionally, the sequence database can be queried based on the antigen sequence to obtain the homologous sequence of the antigen sequence, and the antigen sequence and its homologous sequence are further composed of antigen MSA data and input.

[0398] Specifically, the antibody design model (tFold-Ag) 300 extracts antibody features of the antibody sequence, and the antibody features include: antibody sequence features (s H-L ), antibody pairing characteristics (z H-L ), antibody structure (x H-L ), wherein H represents heavy chain and L represents light chain; and extracting antigenic features of the antigen sequence, the antigenic features including: antigen MSA features (m A ), antigen pairing characteristics (z A ), antigen structure (x A ), where A represents an antigen. The antibody feature and the antigen feature are fused to obtain a fusion feature, which includes: fusion sequence feature (s H-L:A ), fusion pairing features (z H-L:A ); perform complex structure prediction based on fusion features to obtain the predicted complex structure (x H-L:A ), and also generates corresponding confidence scores: plddt = 0.871, ptm = 0.880, iptm = 0.898. The complex loss function of the complex structure prediction task is obtained according to the predicted complex structure and the sample complex structure, and the model parameters are updated. The formula reference is expressed as follows:

[0399] Extraction of antibody sequence features:

[0400] h Ab =tFold-Ab(x Ab )

[0401] Among them, x Ab is the antibody sequence, h Ab is the extracted antibody sequence feature.

[0402] Extraction of initial antibody structure:

[0403] R Ab =Structure(h Ab )

[0404] Among them, R Ab It is the initial antibody structure of the antibody.

[0405] Extraction of antigen sequence features:

[0406] h Ag =Alphafold2(x Ag )

[0407] Among them, x Ag is the antigen sequence, h Ag It is the antigen sequence characteristic.

[0408] Extraction of initial antigen structure:

[0409] R Ag =Structure(h Ag )

[0410] Among them, R Ag It is the initial antigenic structure of the antigen.

[0411] Feature Fusion:

[0412] h fusion =Fusion(h Ab ,h Ag )

[0413] Among them, h fusion It is a fusion feature. The code representation can be specifically referred to the aforementioned method embodiment.

[0414] Structure prediction: Use the Evoformer module to update the fusion features and then perform mapping:

[0415] h′=Evoformer(h fusion )

[0416] Among them, h′ is the fusion feature updated by the Evoformer module and is used to predict the structure of the complex:

[0417] r complex =StructureModule(h′)

[0418] Among them, R complex is the predicted structure of the antibody-antigen complex.

[0419] 2.2: Collaborative tasks between antibody sequence recovery and antibody structure prediction

[0420] Figure 8 It is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application.

[0421] The input data of the antibody design model (tFold-Ag) 300 are: masked antibody sequence (AntibodySequence Masked), antigen sequence (Antigen Sequence), and optionally also include: interaction (PPI) information, epitope (epitope) information, contact (contact) information; the output data are: predicted antibody sequence, predicted complex structure and its confidence score. Among them, the masking technology is used to mask the amino acid sequence corresponding to the complementary determining region in the antibody sequence to obtain the masked antibody sequence, and the predicted antibody sequence is obtained by restoring the amino acid sequence corresponding to the complementary determining region in these masked antibody sequences.

[0422] It should also be noted that when the protein design model executes 2.1 and 2.2 at the same time, the masked antibody sequence used in 2.2 can also be obtained by masking the amino acid sequence corresponding to the complementary determining region in the complete antibody sequence used in 2.1. In this way, the antibody features corresponding to the complete antibody sequence in 2.1 can be fully and effectively utilized in 2.2, which is beneficial to the antibody sequence recovery task.

[0423] Specifically, the antibody design model (tFold-Ag) 300 can perform complex structure prediction based on fusion features to obtain a predicted complex structure, and also perform antibody sequence recovery based on fusion features to obtain a predicted antibody sequence, and also perform antibody structure prediction based on the predicted antibody sequence to obtain a predicted antibody structure, and also generate corresponding confidence scores: iplddt = 0.706, iptm = 0.892. According to the predicted antibody structure and sample antibody structure, the predicted antibody sequence and the sample antibody sequence, the sequence loss function and the structure loss function are obtained and the model parameters are updated. The formula reference is expressed as follows:

[0424] Antibody sequence mask:

[0425] X masker =mask(X Ab )

[0426] Among them, X Ab is the input antibody sequence, X masker is the masked antibody sequence.

[0427] Antibody sequence recovery:

[0428] X′ Ab =SequenceRecovery(X masked ,h Ag )

[0429] Among them, X′ Ab is the recovered predicted antibody sequence, h Ag It is an antigenic characteristic.

[0430] Antibody structure prediction:

[0431] R Ab =StructureModule(X′ Ab ,h Ag )

[0432] Among them, R Ab Based on the predicted antibody sequence X′ Ab Generated predicted antibody structures.

[0433] The loss function of collaborative tasks:

[0434] L=λ struture ·L struture +λ sequence ·L sequence

[0435] Among them, L struture is the structural loss function, L sequence is the sequence loss function, λ struture is the weight of the structural loss function, λ sequence is the weight of the sequence loss function. Through this loss function, the antibody design model (tFold-Ag) 300 can ensure the rationality of its structure and the feasibility of its biological function while generating new antibody sequences.

[0436] (3) Application stage

[0437] The antibody design model (tFold-Ag) 300 is used to perform the coordinated design of antibody sequence and antibody structure, which can realize an antibody having the ability to bind to the antigen indicated by the antigen sequence based on the antigen sequence.

[0438] In certain embodiments, because antibody framework (Antibody Framework Template) is crucial for the stability and solubility of antibody, it can be applied to predict the amino acid sequence corresponding to the complementary determining region in the antibody sequence, and the antibody framework is usually conservative, with only limited variation. In the present embodiment, an antibody framework library is pre-constructed. Specifically, the antibodies in the OAS database are encoded using the Chothia encoding method to obtain the encoded antibody sequence. The encoded antibody sequence is compared with the IMGT standard library, the VJ gene corresponding to the encoded antibody sequence is obtained and its frequency is recorded, the top 2500 antibody sequences with the highest frequency are selected, and the amino acid sequence corresponding to the complementary determining region of these antibody sequences is masked, so as to determine the antibody framework, and form the antibody framework library. Wherein, the antibody sequence is converted to VJ gene, the artificially modified amino acid mutation site can be ignored, and similar homologous antibodies are merged to ensure the versatility of the antibody framework.

[0439] Fig. 9 It is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application. Fig.10 It is a schematic diagram of a protein design method provided by an exemplary embodiment of the present application.

[0440] By inputting these antibody frameworks into the antibody design model (tFold-Ag) 300, a complete predicted antibody sequence and predicted antibody structure can be predicted for each antibody framework. Next, the candidate antibody sequences can be sorted according to the confidence score to facilitate subsequent iterative predictions. In this embodiment, the number of iterations is set to 3 times, that is, antibody design includes three stages:

[0441] In the first stage, all antibody frameworks are obtained from the antibody framework library. These antibody frameworks are obtained by masking the amino acid sequences corresponding to the CDR1, CDR2 and CDR3 regions of the antibody sequence. The antibody sequence is recovered using the antibody design model (tFold-Ag) 300 to obtain the predicted antibody sequence and its predicted antibody structure and confidence score. The predicted antibody sequence with a confidence score higher than the threshold is regarded as a candidate antibody sequence and will enter the second stage.

[0442] In the second stage, the amino acid sequences corresponding to the CDR2 and CDR3 regions in the masked candidate antibody sequences are used to obtain the antibody framework, and the antibody design model (tFold-Ag) 300 is used to perform antibody sequence recovery to obtain the predicted antibody sequence and its predicted antibody structure and confidence score. The predicted antibody sequence with a confidence score higher than the threshold is used as the candidate antibody sequence and will enter the third stage.

[0443] In the third stage, only the amino acid sequence corresponding to the CDR3 region in the candidate antibody series is masked to obtain the antibody framework, and the antibody sequence is recovered using the antibody design model (tFold-Ag) 300 to obtain the predicted antibody sequence and its predicted antibody structure and confidence score. The predicted antibody sequence and predicted antibody structure with the highest confidence can be selected as the final output of the antibody design model (tFold-Ag) 300.

[0444] Optionally, the antibody framework can also be divided into multiple framework clusters, each framework cluster includes multiple antibody frameworks, and the steps of the aforementioned stages are performed on the antibody frameworks in each framework cluster, and the predicted antibody sequence with the highest confidence score is selected from each framework cluster as the representative of the framework cluster. Finally, molecular dynamics simulation can also be used to evaluate the efficacy and stability of the predicted antibody sequence and predicted antibody structure with the highest confidence score. In this embodiment, by using the antibody framework to perform antibody sequence recovery and antibody structure prediction, the accuracy of antibody design can be improved, and biocompatibility and manufacturability can be enhanced.

[0445] (III) Beneficial effects

[0446] The antibody design model (tFold-Ag) is a multi-task model that can not only calculate the antibody-antigen complex structure, but also determine the updated structure of the antibody itself and the antibody structure based on the antibody-antigen complex structure, and can also realize antibody sequence recovery and antibody structure prediction.

[0447] Referring to Table 1, in the performance evaluation, the antibody design model (tFold-Ag) of this embodiment is compared with the end-to-end methods in the related art: AlphaFold-Multimer, etc., and docking-based methods: ZDock, ClusPro, etc. The results show that the antibody design model (tFold-Ag) of this embodiment performs superiorly in the prediction of the antibody-antigen complex structure, with a DockQ score of 0.217 and a TM-score of 0.708. In comparison, AlphaFold-Multimer has a DockQ score of 0.158 and a TM-score of 0.665. In addition, the antibody design model (tFold-Ag) also performs better in the success rate (SR) of antibody-antigen complex prediction, reaching 0.283, while AlphaFold-Multimer is only 0.182.

[0448] Table 1. Comparison of the quality of antibody-antigen complexes predicted by different models

[0449]

[0450]

[0451] Referring to Table 2, in terms of recovery rate, the antibody design model (tFold-Ag) of this embodiment still achieved good results. Computer model evaluation was performed using the amino acid recovery rate (AAR) and contact amino acid recovery rate (CAAR) indicators. In SAbDab-22-DesignAb, the antibody design model (tFold-Ag) of this embodiment achieved AARs of 0.309 and 0.267 for CDR-H3, respectively, without using any prior information. This recovery rate is comparable to the performance of the dyMEAN method, but dyMEAN needs to include prior epitope information. When evaluating the recovery rate of contact residues, the CAAR of the antibody design model (tFold-Ag) of this embodiment was 0.212. The effect of using the interchain features of the antibody design model (tFold-Ag) of this embodiment in antibody design was also evaluated. Combining PPI information can significantly improve design performance. Fig.11 is a distribution diagram of three rounds of ipTM scores provided by an exemplary embodiment of the present application. Fig.11 In the full-length antibody design, the antibody design model (tFold-Ag) screens out unreasonable framework regions after three rounds of iterations. As the number of iterations increases, complex structures with higher confidence scores can be obtained.

[0452] Table 2. Comparison of antibody CDR region recovery rates in different model designs

[0453]

[0454] Referring to Table 3, among the 50 antibodies designed in the laboratory, the antibody design model (tFold-Ag) of this embodiment was used to successfully generate and express these 50 antibodies. In contrast, the expression success rate of the EBM model of the related art was only 42%, and the expression success rate of the dWJS model was 97%.

[0455] Table 3. Comparison of expression success rates of wet experiments of antibodies designed in different models

[0456] method <![CDATA[Expression success rate Total expressed > tFold-Ag 1 wxya 0.97 EBM 0.42 Random 0

[0457] Fig.12 800 is a block diagram of a protein design apparatus provided by an exemplary embodiment of the present application. Exemplarily, the protein design apparatus 800 includes: at least part of the acquisition module 820, the fusion module 840 and the processing module 860.

[0458] An acquisition module 820 is used to acquire a protein framework and extract protein features of the protein framework;

[0459] The acquisition module 820 is used to acquire a target protein sequence and extract target protein features of the target protein sequence;

[0460] A fusion module 840 is used to fuse the protein feature with the target protein feature to obtain a fusion feature; the fusion feature is used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence;

[0461] The processing module 860 is used to perform protein sequence recovery on the protein framework based on the fusion feature to recover the predicted protein sequence corresponding to the protein framework, and to perform protein structure prediction on the predicted protein sequence to obtain the predicted protein structure corresponding to the predicted protein sequence.

[0462] In some embodiments, the protein features include: protein sequence features, protein pairing features and protein structure, the target protein features include: target protein MSA features, target protein pairing features and target protein structure, and the fusion features include: fusion sequence features and fusion pairing features;

[0463] The fusion module 840 is used to:

[0464] Fusion of the protein sequence feature and the target protein MSA feature to obtain the fusion sequence feature;

[0465] Based on the protein structure and the target protein structure, the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature are fused to obtain the fusion pairing feature.

[0466] In some embodiments, the fusion module 840 is used to:

[0467] Performing a linear transformation on the protein sequence features to obtain a transformed protein sequence feature; and performing a linear transformation on the first row of target protein MSA features in the target protein MSA features to obtain a transformed first row of target protein MSA features;

[0468] The transformed protein sequence features and the transformed first-row target protein MSA features are fused based on cross attention to obtain the fused sequence features.

[0469] In some embodiments, the fusion module 840 is used to:

[0470] Determining the protein coordinates corresponding to each amino acid residue in the protein structure based on the carbon atom coordinates of each amino acid residue in the protein structure; and determining the target protein coordinates corresponding to each amino acid residue in the target protein structure based on the carbon atom coordinates of each amino acid residue in the target protein structure;

[0471] Calculating the protein pairing distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the protein structure; and calculating the target protein pairing distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the target protein structure;

[0472] Based on the protein pairing distance and the target protein pairing distance, the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature are fused to obtain the fusion pairing feature.

[0473] In some embodiments, the fusion module 840 is used to:

[0474] Mapping is performed based on the protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature; and mapping is performed based on the target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature;

[0475] Calculating the outer product average based on the protein sequence feature and the target protein MSA feature to obtain the third part of the fusion pairing feature; and determining the transposition of the third part of the fusion pairing feature as the fourth part of the fusion pairing feature;

[0476] The first part of the fusion pairing feature, the second part of the fusion pairing feature, the third part of the fusion pairing feature and the fourth part of the fusion pairing feature are spliced ​​to obtain the fusion pairing feature.

[0477] In some embodiments, the fusion module 840 is used to:

[0478] Performing a linear transformation on the protein pairing feature to obtain a transformed protein pairing feature; and performing a linear transformation on the protein pairing distance to obtain a transformed protein pairing distance;

[0479] The transformed protein pairing feature and the transformed protein pairing distance are added to obtain the first part of the fusion pairing feature.

[0480] In some embodiments, the fusion module 840 is used to:

[0481] Performing a linear transformation on the target protein pairing feature to obtain a transformed target protein pairing feature; and performing a linear transformation on the target protein pairing distance to obtain a transformed target protein pairing distance;

[0482] The transformed target protein pairing feature and the transformed target protein pairing distance are added to obtain the second part of the fusion pairing feature.

[0483] In some embodiments, the fusion module 840 is used to:

[0484] The protein sequence features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized protein sequence features; and the target protein MSA features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized target protein MSA features;

[0485] The outer product average of each element in the transformed normalized protein sequence feature and each element in the transformed normalized target protein MSA feature is calculated, and the outer product average is linearly transformed to obtain the third part fusion pairing feature.

[0486] In some embodiments, the fusion module 840 is further configured to:

[0487] In the presence of an interchain feature or an interaction feature between the protein and the target protein, obtaining an updated sequence feature and an updated pairing feature based on the interchain feature or the interaction feature, as well as the protein structure and the target protein structure;

[0488] The fusion sequence feature is updated based on the updated sequence feature to obtain an updated fusion sequence feature; and the fusion pairing feature is updated based on the updated pairing feature to obtain an updated fusion pairing feature.

[0489] In some embodiments, the fusion module 840 is used to:

[0490] In the presence of the interchain feature, determining the interaction feature based on the interchain feature and the interchain connection relationship between the protein structure and the target protein structure;

[0491] or, in the presence of the interaction feature, determining the interchain feature based on the interaction feature and the interchain pseudo-connection relationship between the protein structure and the target protein structure;

[0492] A linear transformation is performed on the interaction features to obtain the updated sequence features; and a linear transformation is performed on the inter-chain features to obtain the updated pairing features.

[0493] In some embodiments, the processing module 860 is further configured to:

[0494] The protein sequence recovery and the protein structure prediction are sequentially performed on the protein framework of the i-th stage, and a candidate protein sequence is screened out from the predicted protein sequence of the i-th stage based on the confidence scores of the predicted protein sequence and the predicted protein structure respectively corresponding to the protein framework of the i-th stage; wherein i is greater than 0;

[0495] Masking is performed on the amino acid sequence corresponding to the complementary determining region in the candidate protein sequence of the i-th stage to obtain the protein framework of the i+1-th stage; the i is updated to i+1, and the steps of successively performing the protein sequence recovery and the protein structure prediction on the protein framework of the i-th stage are repeated until the set number of iterations is reached, thereby obtaining the predicted protein sequence and predicted protein structure with the highest confidence score.

[0496] In some embodiments, the processing module 860 is configured to:

[0497] Encode the protein sequence in the protein sequence database to obtain the encoded protein sequence;

[0498] Compare the encoded protein sequence with a standard database to obtain the variable region-joint region genes and frequencies corresponding to the encoded protein sequence;

[0499] The protein framework is determined based on the encoded protein sequence having a frequency greater than a threshold.

[0500] In some embodiments, a model structure of a protein design model for performing a protein design method comprises:

[0501] A target protein feature generation unit and a protein feature generation unit connected in parallel; a feature fusion unit connected to the target protein feature generation unit and the protein feature generation unit at the same time; a complex structure prediction unit and a protein sequence recovery unit connected to the feature fusion unit, and a protein structure prediction unit connected to the protein sequence recovery unit;

[0502] Among them, the target protein feature generation unit is used to extract the target protein features of the target protein sequence, and is used to extract the sample target protein features of the sample target protein sequence; the protein feature generation unit is used to extract the protein features of the protein framework, and is used to extract the sample protein features of the sample protein; the feature fusion unit is used to fuse the protein features and the target protein features, and is used to fuse the sample protein features and the sample target protein features; the complex structure prediction unit is used to perform complex structure prediction; the protein sequence recovery unit is used to perform protein sequence recovery; the protein structure prediction unit is used to perform protein structure prediction.

[0503] In some embodiments, the acquisition module 820 is further configured to:

[0504] Acquiring a sample protein sequence, and extracting sample protein features of the sample protein sequence;

[0505] Obtaining a sample target protein sequence, and extracting a sample target protein feature of the sample target protein sequence;

[0506] The fusion module 840 is further configured to:

[0507] The sample protein feature and the sample target protein feature are fused to obtain a sample fusion feature; the sample fusion feature is used to characterize the sequence information and structural information under the interaction between the sample protein indicated by the sample protein sequence and the sample target protein indicated by the sample target protein sequence;

[0508] Also included is a training module; a training module for:

[0509] Prediction is performed on the sample fusion feature to obtain a prediction result, and the model parameters of the protein design model are adjusted with the goal of reducing the difference between the prediction result and the sample result.

[0510] In some embodiments, the sample protein features include: sample protein sequence features, sample protein pairing features and sample protein structure, the sample target protein features include: sample target protein MSA features, sample target protein pairing features and sample target protein structure, and the sample fusion features include: sample fusion sequence features and sample fusion pairing features;

[0511] The fusion module 840 is used to:

[0512] Fusion of the sample protein sequence feature and the sample target protein MSA feature to obtain the sample fusion sequence feature;

[0513] Based on the sample protein structure and the sample target protein structure, the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature are fused to obtain the sample fusion pairing feature.

[0514] In some embodiments, the fusion module 840 is used to:

[0515] Performing a linear transformation on the sample protein sequence features to obtain a transformed sample protein sequence feature; and performing a linear transformation on the first row of sample target protein MSA features in the sample target protein MSA features to obtain a transformed first row of sample target protein MSA features;

[0516] Based on cross attention, the transformed sample protein sequence features and the transformed first-row sample target protein MSA features are fused to obtain the sample fusion sequence features.

[0517] In some embodiments, the fusion module 840 is used to:

[0518] Determining the protein coordinates corresponding to each amino acid residue in the sample protein structure based on the carbon atom coordinates of each amino acid residue in the sample protein structure; and determining the target protein coordinates corresponding to each amino acid residue in the sample target protein structure based on the carbon atom coordinates of each amino acid residue in the sample target protein structure;

[0519] Calculating the protein pairing distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample protein structure; and calculating the target protein pairing distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the sample target protein structure;

[0520] Based on the protein pairing distance and the target protein pairing distance, the sample protein pairing feature, the sample target protein pairing feature, the sample protein sequence feature and the sample target protein MSA feature are fused to obtain the sample fusion pairing feature.

[0521] In some embodiments, the fusion module 840 is used to:

[0522] Mapping is performed based on the sample protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature; and mapping is performed based on the sample target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature;

[0523] Calculating the outer product average based on the sample protein sequence feature and the sample target protein MSA feature to obtain the third part of the fusion pairing feature; and determining the transposition of the third part of the fusion pairing feature as the fourth part of the fusion pairing feature;

[0524] The first part of the fusion pairing features, the second part of the fusion pairing features, the third part of the fusion pairing features and the fourth part of the fusion pairing features are spliced ​​to obtain the sample fusion pairing features.

[0525] In some embodiments, the fusion module 840 is used to:

[0526] Performing a linear transformation on the sample protein pairing feature to obtain a transformed sample protein pairing feature; and performing a linear transformation on the protein pairing distance to obtain a transformed protein pairing distance;

[0527] The transformed sample protein pairing feature and the transformed protein pairing distance are added to obtain the first part of the fusion pairing feature.

[0528] In some embodiments, the fusion module 840 is used to:

[0529] Performing a linear transformation on the sample target protein pairing feature to obtain a transformed sample target protein pairing feature; and performing a linear transformation on the target protein pairing distance to obtain a transformed target protein pairing distance;

[0530] The transformed sample target protein pairing feature and the transformed target protein pairing distance are added to obtain the second part of the fusion pairing feature.

[0531] In some embodiments, the fusion module 840 is used to:

[0532] The sample protein sequence features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized protein sequence features; and the sample target protein MSA features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized target protein MSA features;

[0533] The outer product average of each element in the transformed normalized protein sequence feature and each element in the transformed normalized target protein MSA feature is calculated, and the outer product average is linearly transformed to obtain the third part fusion pairing feature.

[0534] In some embodiments, the fusion module 840 is further configured to:

[0535] In the presence of a sample interchain feature or a sample interaction feature between the sample protein and the sample target protein, a sample updated sequence feature and a sample updated pairing feature are obtained based on the sample interchain feature or the sample interaction feature, and the sample protein structure and the sample target protein structure;

[0536] The sample fusion sequence feature is updated based on the sample update sequence feature to obtain an updated sample fusion sequence feature; and the sample fusion pairing feature is updated based on the sample update pairing feature to obtain an updated sample fusion pairing feature.

[0537] In some embodiments, the fusion module 840 is used to:

[0538] In the case where the sample inter-chain feature exists, the sample interaction feature is determined based on the sample inter-chain feature and the inter-chain connection relationship; or, in the case where the sample interaction feature exists, the sample inter-chain feature is determined based on the sample interaction feature and the inter-chain pseudo connection relationship;

[0539] Performing a linear transformation on the sample interaction feature to obtain the sample update sequence feature; and performing a linear transformation on the sample inter-chain feature to obtain the sample update pairing feature.

[0540] In some embodiments, the prediction is a multi-task prediction, the prediction results include: predicted complex structure, predicted protein sequence and predicted protein structure, and the sample results include: sample complex structure, sample protein sequence and sample protein structure;

[0541] Training modules for:

[0542] Performing complex structure prediction on the sample fusion features to obtain the predicted complex structure;

[0543] Masking the amino acid sequence corresponding to the complementarity determining region in the sample protein sequence to obtain a masked sample protein sequence;

[0544] Performing protein sequence recovery on the sample fusion feature to recover the predicted protein sequence corresponding to the masked sample protein sequence, and performing protein structure prediction on the predicted protein sequence to obtain the predicted protein structure;

[0545] The model parameters of the protein design model are adjusted with the training goal of reducing the difference between the predicted complex structure and the sample complex structure, the difference between the predicted protein sequence and the sample protein sequence, and the difference between the predicted protein structure and the sample protein structure.

[0546] In some embodiments, the training module is used to:

[0547] Based on the predicted complex structure and the sample complex structure, determining a complex loss function, wherein the complex loss function is obtained by performing a weighted summation on at least one of the following losses: a geometric loss between amino acid residues for reducing the distance and angle between amino acid residues, a frame alignment point error loss for preventing local structure destruction, an interface frame alignment point error loss for constraining inter-chain amino acid residue pairs, a coordinate root mean square difference loss for aligning coordinates, a confidence loss for estimating a confidence score, and a structural violation loss for handling structural violations;

[0548] Determining a structure loss function based on the predicted protein structure and the sample protein structure, and determining a sequence loss function based on the predicted protein sequence and the sample protein sequence;

[0549] The complex loss function, the structure loss function and the sequence loss function are weightedly summed to obtain a total loss function; and the model parameters of the protein design model are adjusted with reducing the total loss function as the training goal.

[0550] It should be noted that the specific definitions in the embodiments of the one or more protein design devices 800 provided above can refer to the definitions of the protein design method above, and will not be repeated here. Each module of the above device can be implemented in whole or in part by software, hardware, or a combination thereof. Each module can be embedded in or independent of a processor of a computer device in the form of hardware, or can be stored in a memory of a computer device in the form of software, so that the processor can call and execute operations corresponding to each module.

[0551] An embodiment of the present application also provides a computer device, which includes: a processor and a memory, wherein a computer program is stored in the memory; and the processor is used to execute the computer program in the memory to implement the protein design method provided by each of the above method embodiments.

[0552] For example, Fig.13 1 is a structural block diagram of a computer device provided by an exemplary embodiment of the present application. Optionally, the computer device is a server 1000.

[0553] Typically, the server 1000 includes: a processor 1001 and a memory 1002 .

[0554] The processor 1001 may include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 1001 may be implemented in at least one hardware form of digital signal processing (DSP), field programmable gate array (FPGA), and programmable logic array (PLA). The processor 1001 may also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also known as a central processing unit (CPU); the coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 1001 may be integrated with a graphics processing unit (GPU), which is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 1001 may also include an artificial intelligence (AI) processor, which is used to process computing operations related to machine learning.

[0555] The memory 1002 may include one or more computer-readable storage media, which may be non-transitory. The memory 1002 may also include a high-speed random access memory, and a non-volatile memory, such as one or more disk storage devices, flash memory storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 1002 is used to store at least one instruction, which is used to be executed by the processor 1001 to implement the protein design method provided by each method embodiment of the present application.

[0556] In some embodiments, the server 1000 may also optionally include: an input interface 1003 and an output interface 1004. The processor 1001, the memory 1002, the input interface 1003, and the output interface 1004 may be connected via a bus or a signal line. Each peripheral device may be connected to the input interface 1003 and the output interface 1004 via a bus, a signal line, or a circuit board. The input interface 1003 and the output interface 1004 may be used to connect at least one peripheral device related to input / output (I / O) to the processor 1001 and the memory 1002. In some embodiments, the processor 1001, the memory 1002, the input interface 1003, and the output interface 1004 are integrated on the same chip or circuit board; in some other embodiments, any one or two of the processor 1001, the memory 1002, the input interface 1003, and the output interface 1004 may be implemented on a separate chip or circuit board, which is not limited in the embodiments of the present application.

[0557] Those skilled in the art will understand that Fig.13 The structure shown in the figure does not constitute a limitation on the computer device, and may include more or less components than shown in the figure, or combine some components, or adopt a different component arrangement.

[0558] In an exemplary embodiment, the present application also provides a chip, which includes a programmable logic circuit and / or program instructions. When the chip runs on a computer device, it is used to implement the protein design method provided by the above method embodiment.

[0559] An embodiment of the present application also provides a computer-readable storage medium, which stores a computer program. The computer program is loaded and executed by a processor to implement the protein design method provided by the above method embodiment.

[0560] The present application embodiment also provides a computer program product or computer program, the computer program product or computer program includes computer instructions, the computer instructions are stored in a computer-readable storage medium. The processor of the computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the processor of the computer device loads and executes to implement the protein design method provided by the above method embodiment.

[0561] The serial numbers of the above-mentioned embodiments of the present application are for description only and do not represent the advantages or disadvantages of the embodiments.

[0562] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware or by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, and the above-mentioned computer-readable storage medium may be a read-only memory, a disk or an optical disk, etc.

[0563] Those skilled in the art should be aware that in one or more of the above examples, the functions described in the embodiments of the present application can be implemented with hardware, software, firmware, or any combination thereof. When implemented using software, these functions can be stored in a computer-readable medium or transmitted as one or more instructions or codes on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein the communication media include any media that facilitates the transmission of a computer program from one place to another. The storage medium can be any available medium that a general or special-purpose computer can access.

[0564] The above description is only an optional embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A protein design method, characterized in that: The method comprises: Acquiring a protein framework, and extracting protein features of the protein framework; Obtaining a target protein sequence, and extracting target protein features of the target protein sequence; The protein feature and the target protein feature are fused to obtain a fusion feature; the fusion feature is used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence; Based on the fusion feature, protein sequence recovery is performed on the protein framework to recover the predicted protein sequence corresponding to the protein framework, and protein structure prediction is performed on the predicted protein sequence to obtain the predicted protein structure corresponding to the predicted protein sequence.

2. The method according to claim 1, characterized in that The protein features include: protein sequence features, protein pairing features and protein structure; the target protein features include: target protein MSA features, target protein pairing features and target protein structure; the fusion features include: fusion sequence features and fusion pairing features; The step of fusing the protein feature with the target protein feature to obtain a fusion feature comprises: Fusion of the protein sequence feature and the target protein MSA feature to obtain the fusion sequence feature; Based on the protein structure and the target protein structure, the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature are fused to obtain the fusion pairing feature.

3. The method according to claim 2, characterized in that The step of fusing the protein sequence feature and the target protein MSA feature to obtain the fusion sequence feature comprises: Performing a linear transformation on the protein sequence features to obtain a transformed protein sequence feature; and performing a linear transformation on the first row of target protein MSA features in the target protein MSA features to obtain a transformed first row of target protein MSA features; The transformed protein sequence features and the transformed first-row target protein MSA features are fused based on cross attention to obtain the fused sequence features.

4. The method according to claim 2, characterized in that: The method of obtaining the fusion pairing feature by fusing the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature based on the protein structure and the target protein structure includes: Determining the protein coordinates corresponding to each amino acid residue in the protein structure based on the carbon atom coordinates of each amino acid residue in the protein structure; and determining the target protein coordinates corresponding to each amino acid residue in the target protein structure based on the carbon atom coordinates of each amino acid residue in the target protein structure; Calculating the protein pairing distance between the protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the protein structure; and calculating the target protein pairing distance between the target protein coordinates corresponding to any two amino acid residues located in the same peptide chain in the target protein structure; Based on the protein pairing distance and the target protein pairing distance, the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature are fused to obtain the fusion pairing feature.

5. The method according to claim 4, characterized in that The step of fusing the protein pairing feature, the target protein pairing feature, the protein sequence feature and the target protein MSA feature based on the protein pairing distance and the target protein pairing distance to obtain the fusion pairing feature comprises: Mapping is performed based on the protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature; and mapping is performed based on the target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature; Calculating the outer product average based on the protein sequence feature and the target protein MSA feature to obtain the third part of the fusion pairing feature; and determining the transposition of the third part of the fusion pairing feature as the fourth part of the fusion pairing feature; The first part of the fusion pairing feature, the second part of the fusion pairing feature, the third part of the fusion pairing feature and the fourth part of the fusion pairing feature are spliced ​​to obtain the fusion pairing feature.

6. The method according to claim 5, characterized in that The performing of mapping based on the protein pairing feature and the protein pairing distance to calculate the first part of the fusion pairing feature comprises: Performing a linear transformation on the protein pairing feature to obtain a transformed protein pairing feature; and performing a linear transformation on the protein pairing distance to obtain a transformed protein pairing distance; The transformed protein pairing feature and the transformed protein pairing distance are added to obtain the first part of the fusion pairing feature.

7. The method according to claim 5, characterized in that The performing of mapping based on the target protein pairing feature and the target protein pairing distance to calculate the second part of the fusion pairing feature includes: Performing a linear transformation on the target protein pairing feature to obtain a transformed target protein pairing feature; and performing a linear transformation on the target protein pairing distance to obtain a transformed target protein pairing distance; The transformed target protein pairing feature and the transformed target protein pairing distance are added to obtain the second part of the fusion pairing feature.

8. The method according to claim 5, characterized in that The outer product average is calculated based on the protein sequence feature and the target protein MSA feature to obtain the third part of the fusion pairing feature, including: The protein sequence features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized protein sequence features; and the target protein MSA features are sequentially subjected to layer normalization processing and linear transformation to obtain transformed normalized target protein MSA features; The outer product average of each element in the transformed normalized protein sequence feature and each element in the transformed normalized target protein MSA feature is calculated, and the outer product average is linearly transformed to obtain the third part fusion pairing feature.

9. The method according to any one of claims 2 to 8, characterized in that: The method further comprises: In the presence of an interchain feature or an interaction feature between the protein and the target protein, obtaining an updated sequence feature and an updated pairing feature based on the interchain feature or the interaction feature, as well as the protein structure and the target protein structure; The fusion sequence feature is updated based on the updated sequence feature to obtain an updated fusion sequence feature; and the fusion pairing feature is updated based on the updated pairing feature to obtain an updated fusion pairing feature.

10. The method according to claim 9, characterized in that In the case where there are interchain features or interaction features, obtaining updated sequence features and updated pairing features based on the interchain features or the interaction features, as well as the protein structure and the target protein structure, includes: In the presence of the interchain feature, determining the interaction feature based on the interchain feature and the interchain connection relationship between the protein structure and the target protein structure; or, in the presence of the interaction feature, determining the interchain feature based on the interaction feature and the interchain pseudo-connection relationship between the protein structure and the target protein structure; A linear transformation is performed on the interaction features to obtain the updated sequence features; and a linear transformation is performed on the inter-chain features to obtain the updated pairing features.

11. The method according to any one of claims 1 to 10, characterized in that: The method further comprises: The protein sequence recovery and the protein structure prediction are sequentially performed on the protein framework of the i-th stage, and a candidate protein sequence is screened out from the predicted protein sequence of the i-th stage based on the confidence scores of the predicted protein sequence and the predicted protein structure respectively corresponding to the protein framework of the i-th stage; wherein i is greater than 0; Masking is performed on the amino acid sequence corresponding to the complementary determining region in the candidate protein sequence of the i-th stage to obtain the protein framework of the i+1-th stage; the i is updated to i+1, and the steps of successively performing the protein sequence recovery and the protein structure prediction on the protein framework of the i-th stage are repeated until the set number of iterations is reached, thereby obtaining the predicted protein sequence and predicted protein structure with the highest confidence score.

12. The method according to any one of claims 1 to 10, characterized in that: The method further comprises: Encode the protein sequence in the protein sequence database to obtain the encoded protein sequence; Compare the encoded protein sequence with a standard database to obtain the variable region-joint region genes and frequencies corresponding to the encoded protein sequence; The protein framework is determined based on the encoded protein sequence having a frequency greater than a threshold.

13. The method according to any one of claims 1 to 12, characterized in that: Model structures of protein design models used to implement protein design methods, including: A target protein feature generation unit and a protein feature generation unit connected in parallel; a feature fusion unit connected to the target protein feature generation unit and the protein feature generation unit at the same time; a complex structure prediction unit and a protein sequence recovery unit connected to the feature fusion unit, and a protein structure prediction unit connected to the protein sequence recovery unit; Among them, the target protein feature generation unit is used to extract the target protein features of the target protein sequence, and is used to extract the sample target protein features of the sample target protein sequence; the protein feature generation unit is used to extract the protein features of the protein framework, and is used to extract the sample protein features of the sample protein; the feature fusion unit is used to fuse the protein features and the target protein features, and is used to fuse the sample protein features and the sample target protein features; the complex structure prediction unit is used to perform complex structure prediction; the protein sequence recovery unit is used to perform protein sequence recovery; the protein structure prediction unit is used to perform protein structure prediction.

14. The method according to any one of claims 1 to 13, characterized in that: The method further comprises: Acquiring a sample protein sequence, and extracting sample protein features of the sample protein sequence; Obtaining a sample target protein sequence, and extracting a sample target protein feature of the sample target protein sequence; The sample protein feature and the sample target protein feature are fused to obtain a sample fusion feature; the sample fusion feature is used to characterize the sequence information and structural information under the interaction between the sample protein indicated by the sample protein sequence and the sample target protein indicated by the sample target protein sequence; Prediction is performed on the sample fusion feature to obtain a prediction result, and the model parameters of the protein design model are adjusted with the goal of reducing the difference between the prediction result and the sample result.

15. The method according to claim 14, characterized in that The prediction is a multi-task prediction, the prediction results include: predicted complex structure, predicted protein sequence and predicted protein structure, and the sample results include: sample complex structure, sample protein sequence and sample protein structure; The step of performing prediction on the sample fusion feature to obtain a prediction result, taking reducing the difference between the prediction result and the sample result as a training goal, and adjusting the model parameters of the protein design model comprises: Performing complex structure prediction on the sample fusion features to obtain the predicted complex structure; Masking the amino acid sequence corresponding to the complementarity determining region in the sample protein sequence to obtain a masked sample protein sequence; Performing protein sequence recovery on the sample fusion feature to recover the predicted protein sequence corresponding to the masked sample protein sequence, and performing protein structure prediction on the predicted protein sequence to obtain the predicted protein structure; The model parameters of the protein design model are adjusted with the training goal of reducing the difference between the predicted complex structure and the sample complex structure, the difference between the predicted protein sequence and the sample protein sequence, and the difference between the predicted protein structure and the sample protein structure.

16. The method according to claim 15, characterized in that The adjusting the model parameters of the protein design model with the training goal of reducing the difference between the predicted complex structure and the sample complex structure, the difference between the predicted protein sequence and the sample protein sequence, and the difference between the predicted protein structure and the sample protein structure comprises: Based on the predicted complex structure and the sample complex structure, determining a complex loss function, wherein the complex loss function is obtained by performing a weighted summation on at least one of the following losses: a geometric loss between amino acid residues for reducing the distance and angle between amino acid residues, a frame alignment point error loss for preventing local structure destruction, an interface frame alignment point error loss for constraining inter-chain amino acid residue pairs, a coordinate root mean square difference loss for aligning coordinates, a confidence loss for estimating a confidence score, and a structural violation loss for handling structural violations; Determining a structure loss function based on the predicted protein structure and the sample protein structure, and determining a sequence loss function based on the predicted protein sequence and the sample protein sequence; The complex loss function, the structure loss function and the sequence loss function are weightedly summed to obtain a total loss function; and the model parameters of the protein design model are adjusted with reducing the total loss function as the training goal.

17. A protein design device, characterized in that: The device comprises: An acquisition module, used for acquiring a protein framework and extracting protein features of the protein framework; The acquisition module is used to acquire the target protein sequence and extract the target protein features of the target protein sequence; A fusion module, used to fuse the protein feature with the target protein feature to obtain a fusion feature; the fusion feature is used to characterize the sequence information and structural information under the interaction between the protein indicated by the protein framework and the target protein indicated by the target protein sequence; A processing module is used to perform protein sequence recovery on the protein framework based on the fusion feature to recover the predicted protein sequence corresponding to the protein framework, and to perform protein structure prediction on the predicted protein sequence to obtain the predicted protein structure corresponding to the predicted protein sequence.

18. A computer device, characterized in that: The computer device comprises: a processor and a memory, wherein the memory stores a computer program, and the computer program is loaded and executed by the processor to implement the protein design method according to any one of claims 1 to 16.

19. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and the computer program is loaded and executed by a processor to implement the protein design method according to any one of claims 1 to 16.

20. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. The processor obtains the computer instructions from the computer-readable storage medium, so that the processor loads and executes them to implement the protein design method according to any one of claims 1 to 16.

Images