Protein complex structure prediction method and device
The method enhances protein complex structure prediction by using an N-stage folding iterative network layer to incorporate residue and monomer chain-level positional transformations, addressing inefficiencies in existing methods and improving accuracy for protein complexes with multiple chains.
Patent Information
- Application Number
- JP2024145815
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Priority Date
- 2023-11-08
- Filing Date
- 2024-08-27
- Publication Date
- 2024-11-19
- Estimated Expiration
- 2044-08-27
AI Technical Summary
Existing methods for predicting protein complex structures are inefficient and do not adequately account for the relative independence of monomer chains, limiting their applicability to protein complexes with multiple chains.
A method and apparatus that utilize an N-stage folding iterative network layer to predict protein complex structures by incorporating initial coordinates, residue pair features, and multiple sequence alignment features, with positional transformations at both residue and monomer chain levels, enhancing accuracy and efficiency.
The method improves the efficiency and accuracy of protein complex structure prediction, particularly in scenarios involving multiple chains, by accounting for the relative independence of monomer chains and integrating monomer chain-level positional transformations.
Smart Images

Figure 2024161599000001_ABST
Abstract
Description
[Technical field]
[0001] The present disclosure relates to the field of artificial intelligence technology, and in particular to technical fields such as natural language processing and biocomputing. [Background technology]
[0002] A protein complex is a stable macromolecular complex formed by the interaction of two or more protein molecules, and plays an important role in various biological functions such as enzyme reactions, cell signaling, metabolic regulation, gene expression, etc. Here, the function of a protein is largely determined by the spatial structure of the protein itself, and a technique for predicting the three-dimensional structure (tertiary structure) of a protein in space based on the amino acid categories (primary structure) of the protein chain has extremely high research value in the field of life sciences.
[0003] Therefore, how to accurately predict protein structures and improve the efficiency of protein complex structure prediction to meet various biological applications has become an important research direction. Summary of the Invention [Problem to be solved by the invention]
[0004] The present disclosure provides a method and apparatus for predicting the structure of a protein complex. [Means for solving the problem]
[0005] According to one aspect of the present disclosure, there is provided a method for predicting a structure of a protein complex, the method comprising: Obtaining initial coordinates of each amino acid residue in the target protein complex, and obtaining a target residue pair feature, a first multiple sequence alignment feature, and a second multiple sequence alignment feature of each protein monomer in the target protein complex; inputting the initial coordinates of each amino acid residue, the target residue pair feature of each protein monomer, the first multiple sequence alignment feature and the second multiple sequence alignment feature into an N-stage folded repeat network layer, and predicting the torsion angle, residue-level position transformation and monomer chain-level position transformation of each amino acid residue through the N-stage folded repeat network layer to obtain the target coordinates of each amino acid residue, and obtain a predicted structure of the protein complex; Here, the first multiple sequence alignment feature is a regularized multiple sequence alignment feature, the second multiple sequence alignment feature is a mapped multiple sequence alignment feature, and N is an integer greater than 1.
[0006] The present disclosure takes into account the relative independence of each monomer chain in a protein complex, and adds monomer chain level position conversion to the residue level position conversion, thereby updating the coordinates of each amino acid residue, accurately predicting the structure of a predicted protein, improving the efficiency of protein complex structure prediction, and making it better applicable to application scenarios in which a protein complex contains multiple chains.
[0007] According to another aspect of the present disclosure, there is provided a structure prediction apparatus for a protein complex, the apparatus comprising: an acquisition module for acquiring initial coordinates of each amino acid residue in the target protein complex, and acquiring a target residue pair feature, a first multiple sequence alignment feature and a second multiple sequence alignment feature of each protein monomer in the target protein complex; a structure prediction module for inputting the initial coordinates of each amino acid residue, the target residue pair feature of each protein monomer, the first multiple sequence alignment feature and the second multiple sequence alignment feature into an N-stage folded repeat network layer, and predicting the torsion angle, residue-level position transformation and monomer chain-level position transformation of each amino acid residue through the N-stage folded repeat network layer to obtain the target coordinates of each amino acid residue, and obtain a predicted structure of the protein complex; Here, the first multiple sequence alignment feature is a regularized multiple sequence alignment feature, the second multiple sequence alignment feature is a mapped multiple sequence alignment feature, and N is an integer greater than 1.
[0008] According to another aspect of the present disclosure, an electronic device is provided, comprising: at least one processor; and a memory communicatively coupled to the at least one processor, wherein: The memory stores instructions to be executed by at least one processor, and the instructions, when executed by the at least one processor, cause the at least one processor to perform a protein complex structure prediction method according to an embodiment of the first aspect of the present disclosure.
[0009] According to another aspect of the present disclosure, there is provided a non-transitory computer readable storage medium having stored thereon computer instructions, wherein the computer instructions cause a computer to perform a protein complex structure prediction method according to an embodiment of the first aspect of the present disclosure.
[0010] According to another aspect of the present disclosure, there is provided a computer program, which, when executed by a processor, realizes steps of the method for predicting a protein complex structure according to an embodiment of the first aspect of the present disclosure.
[0011] However, the contents described in this section are not intended to identify the essential or important features of the embodiments of the present disclosure, and are not intended to limit the scope of the present disclosure. Other features of the present disclosure will become more readily understood from the following specification. [Brief description of the drawings]
[0012] The drawings are for a better understanding of the present solution and are not intended to limit the present disclosure. [Figure 1] 1 is a flowchart of a method for predicting a structure of a protein complex according to one embodiment of the present disclosure. [Diagram 2]1 is a flowchart of a method for predicting a structure of a protein complex according to one embodiment of the present disclosure. [Diagram 3] FIG. 2 is a structural diagram of a protein complex structure prediction method according to an embodiment of the present disclosure. [Figure 4] 1 is a flowchart of a method for predicting a structure of a protein complex according to one embodiment of the present disclosure. [Diagram 5] FIG. 2 is a structural diagram of a protein complex structure prediction method according to an embodiment of the present disclosure. [Figure 6] FIG. 1 is a structural diagram of a protein complex structure prediction device according to an embodiment of the present disclosure. [Figure 7] FIG. 1 is a block diagram of an electronic device for implementing a method according to an embodiment of the present disclosure. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS
[0013] In the following, exemplary embodiments of the present disclosure will be described in conjunction with the drawings, and various details of the embodiments of the present disclosure will be included for ease of understanding, and they should be regarded as illustrative. Therefore, as will be understood by those skilled in the art, various changes and modifications can be made to the embodiments without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, the following description omits descriptions of well-known functions and structures.
[0014] Embodiments of the present disclosure relate to the field of artificial intelligence techniques such as computer vision and deep learning.
[0015] Artificial Intelligence, abbreviated as AI in English, is a technical science that researches and develops theories, methods, technologies and application systems that simulate and extend human intelligence.
[0016] Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various logics and methods to realize effective communication in natural language between humans and computers. NLP is a science that integrates linguistics, computer science and mathematics. Therefore, this field of research is concerned with natural language, i.e. the language that people use in daily life, and is therefore closely related to the study of linguistics, but with important differences. NLP is not the study of natural language in general, but the study of computer systems, especially software systems, that can realize natural language communication effectively. It is a part of computer science.
[0017] Biocomputing refers to a new computing mode that utilizes the inherent information processing mechanisms of biological systems. Biocomputing research includes two aspects: devices and systems. It uses ordered systems constructed on a molecular scale using organic (or biological) materials to provide basic units that detect, process, transmit, and store information through physical and chemical processes at the molecular level.
[0018] The method and apparatus for predicting a protein complex structure according to the present disclosure will be described below in conjunction with the drawings.
[0019] FIG. 1 is a flowchart of a method for predicting a protein complex structure according to one embodiment of the present disclosure. As shown in FIG. 1, the method includes the following steps S101 to S102.
[0020] In S101, the initial coordinates of each amino acid residue in the target protein complex are obtained, and a target residue pair feature, a first multiple sequence alignment feature and a second multiple sequence alignment feature of each protein monomer in the target protein complex are obtained.
[0021] A protein complex has multiple protein monomers, each of which has an amino acid sequence, and the amino acids lose one water molecule when they bind to each other to form a peptide bond, so that the amino acid unit in a polypeptide / protein is an amino acid residue. In the embodiment of the present disclosure, in order to accommodate the rotational invariance of the protein structure, a relative position transformation is used to represent the coordinates of each residue, and the spatial structure of the protein complex is initialized at the origin of the coordinates, that is, the coordinates of each amino acid residue in the target protein complex are initialized with the initial coordinate T i = (I,0(→)), where T i =(I,0(→)) represents the apoapsis coordinates expressed by rotation / translation, I is the identity matrix indicating no rotation, the 0(→) vector indicates no translation, and i represents the i-th amino acid residue.
[0022] In an embodiment of the present disclosure, a template feature is obtained for each of the protein monomers, a pair feature of the amino acid sequence of each of the protein monomers is constructed, and for each protein monomer, a target residue pair feature of the protein monomer is obtained based on the template feature and pair feature of the protein monomer.
[0023] In some embodiments, for each protein monomer, a homologous sequence of the protein monomer is searched and obtained from multiple gene sequence databases based on the target amino acid sequence of the protein monomer, a multiple sequence alignment is performed on the homologous sequence of the protein monomer to obtain a multiple sequence alignment feature of the protein monomer, and further, based on the multiple sequence alignment feature, different processing is performed to obtain a first multiple sequence alignment feature and a second multiple sequence alignment feature, and optionally, the first multiple sequence alignment feature is a regularized multiple sequence alignment feature and the second multiple sequence alignment feature is a mapped multiple sequence alignment feature.
[0024] In S102, the initial coordinates of each amino acid residue, the target residue pair feature of each protein monomer, the first multiple sequence alignment feature and the second multiple sequence alignment feature are input into an N-stage folded repeat network layer, and the N-stage folded repeat network layer predicts the torsion angle, residue-level position transformation and monomer chain-level position transformation of each amino acid residue to obtain the target coordinates of each amino acid residue, and obtain a predicted structure of the protein complex, where N is an integer greater than 1.
[0025] Optionally, torsion angles at residue side chains can be predicted based on a side chain and torsion angle predictor at the N-stage folded repeat network layer.
[0026] In some embodiments, protein complexes are predicted by directly mapping the residue codes of multiple chains in the complex to coordinate transformations, where these transformations operate only on residues, and the present disclosure refers to such transformations as residue-level positional transformations.
[0027] In the embodiment of the present disclosure, the relative independence of each monomer chain in a protein complex is taken into account, and the coordinates of each amino acid residue are updated by adding a monomer chain-level position conversion to the residue-level position conversion, thereby realizing a separation between intra-chain residue position prediction and daughter chain overall position prediction, and improving the overall effectiveness of the protein structure prediction model.
[0028] In an embodiment of the present disclosure, the initial coordinates of each amino acid residue, the target residue pair feature of each protein monomer, the first multiple sequence alignment feature and the second multiple sequence alignment feature are input into an N-stage folded repeat network layer, and the torsion angle, residue-level position transformation and monomer chain-level position transformation of each amino acid residue are predicted by the N-stage folded repeat network layer to obtain the target coordinates of each amino acid residue, thereby taking into account the relative independence of each monomer chain in the protein complex, and adding the monomer chain-level position transformation to the base of the residue-level position transformation to update the coordinates of each amino acid residue, thereby accurately predicting the structure of the predicted protein, improving the efficiency of protein complex structure prediction, and better applicable to application scenarios in which the protein complex contains multiple chains.
[0029] FIG. 2 is a flowchart of a method for predicting a protein complex structure according to one embodiment of the present disclosure. As shown in FIG. 2, the method includes the following steps S201 to S204.
[0030] In S201, the initial coordinates of each amino acid residue in the target protein complex are obtained, and the target residue pair feature, the first multiple sequence alignment feature and the second multiple sequence alignment feature of each protein monomer in the target protein complex are obtained.
[0031] For the introduction of step S201, please refer to the relevant explanation in the above embodiment, and detailed explanation will be omitted here.
[0032] In S202, the initial coordinates, the target residue pair features and the second multiple sequence alignment features are input into the first-stage folded repeat network layer, and residue-level position transformation predictions and monomer-chain-level position transformation predictions are performed for each amino acid residue to obtain the target residue code 1 and candidate position transformation 1 of the first-stage folded repeat network layer.
[0033] The first-stage folding iterative network layer performs an invariant point attention mechanism on the initial coordinates, the target residue pair features and the second multiple sequence alignment features to obtain a residue code with rotational invariance, and then performs a mapping process using a linear network to obtain the target residue code 1.
[0034] Based on the backbone update algorithm, a target residue code of each amino acid residue is mapped, and a residue-level positional transformation prediction is performed for the target residue code 1 to obtain a first positional transformation 1 of each amino acid residue, and a monomer chain-level positional transformation prediction is performed for the target residue code 1 to obtain a second positional transformation 1 of each amino acid residue.
[0035] As shown in FIG. 3 , in some embodiments, the monomer chain-level Chain Affine Update process includes, for each amino acid residue, splitting two or more adjacent amino acid residues into different monomer chains based on the target residue code of the amino acid residue, for example, splitting three adjacent amino acid residues into the same monomer chain, i.e., the residue codes [s1, s2, s3, ...., s i ,….,s r ] and the monomer chain before splicing (for example, s1 to s3 belong to monomer chain 1 and s r-2 ~s r belongs to monomer chain n) and calculate the average of the target residue codes for each monomer chain to obtain a chain-level representation of the candidate residue codes.
[0036] In an embodiment of the present disclosure, for target amino acid residues in any one monomer chain, the average value (mean) of the target residue codes of the target amino acid residues is calculated to obtain candidate residue codes at the chain level, and the candidate residue codes are mapped based on a multi-layer neural network structure (the multi-layer linear network Linear shown in FIG. 3 ) to obtain a second position transformation of each amino acid residue in the monomer chain.
[0037] As shown in Figure 3, in some embodiments, the multi-layer neural network structure includes a three-layer linear network structure, in which the candidate residue code is input to a first linear network for mapping to obtain a first transformed representation. The first transformed representation is input to a second linear network for mapping to obtain a second transformed representation. The first transformed representation and the second transformed representation are input to a third linear network for mapping to obtain a second position transformation of each amino acid residue in the monomer chain. The structures of the three-layer linear networks may be the same or different, and the embodiments of the present application are not limited thereto.
[0038] A position update is performed based on the first position transformation 1, the second position transformation 1, and the initial coordinates to obtain a candidate position transformation 1 for the first stage folding network layer.
[0039] In S203, for the m-th stage folded repeat network layer, the target residue pair feature, the target residue code m-1 of the m-1-th stage folded repeat network layer, and the candidate position transformation m-1 are input into the m-th stage folded repeat network layer to perform residue level position transformation prediction and monomer chain level position transformation prediction for each amino acid residue, thereby obtaining the target residue code m and candidate position transformation m of the m-th stage folded repeat network layer, where the value of m is 2 to N.
[0040] An invariant attention mechanism is performed on the candidate position transformation m-1 of the m-1-th folding iterative network layer input into the m-th folding network layer and the target residue pair feature to obtain a residue code with rotation invariance, and a mapping process is performed by a linear network to obtain the target residue code m.
[0041] Similarly, using the method in step S202, continue to predict residue-level position transformations for the target residue code m to obtain a first position transformation m for each amino acid residue, and predict monomer-chain-level position transformations for the target residue code m to obtain a second position transformation m for each amino acid residue, until obtaining a candidate position transformation N and a target residue code N for the Nth stage of the folding network layer, and obtain a candidate position transformation m for the mth stage of the folding network layer based on the first position transformation m and the second position transformation m.
[0042] In S204, the Nth stage folded repeat network layer performs side chain and torsion angle prediction for the first multiple sequence alignment feature and the target residue code N of the Nth stage folded repeat network layer to obtain the torsion angle of the side chain of each amino acid residue, and obtains the target coordinate of each amino acid residue based on the torsion angle of the side chain of each amino acid residue and the candidate position transformation N of the Nth stage folded repeat network layer.
[0043] The first multiple sequence alignment feature and the target residue code N of the Nth stage folded repeat network layer are input into a side chain and torsion angle predictor of the Nth stage folded repeat network layer to obtain the torsion angles of the side chains of the amino acid residues, and the target coordinates of the amino acid residues are obtained based on the torsion angles of the side chains of each amino acid residue, the candidate position transformation N of the Nth stage folded repeat network layer and the backward position update.
[0044] In the embodiment of the present application, the relative independence of each monomer chain in a protein complex is taken into account, and by adding monomer chain-level position transformation to the residue-level position transformation, the coordinates of each amino acid residue can be updated and separation of intra-chain residue position prediction and daughter chain overall position prediction can be realized, which improves the overall effect of the protein structure prediction model, and while maintaining the relative positions of residues in a single chain, the docking relationship between chains can be more appropriately adjusted overall, which can better fit the protein complex structure prediction.
[0045] FIG. 4 is a flowchart of a method for predicting a protein complex structure according to one embodiment of the present disclosure. As shown in FIG. 4, the method includes the following steps S401 to S408.
[0046] In S401, a template feature of each protein monomer is obtained, and a pair feature of the amino acid sequence of each protein monomer is constructed.
[0047] In some embodiments, the target amino acid sequence of each protein monomer is matched against a plurality of first amino acid sequences in a protein structure database to obtain a second amino acid sequence having a similarity greater than a preset threshold, and the distance between the coordinates of the amino acid residues in the second amino acid sequence is extracted as a template feature of each protein monomer. That is, a protein structure having a similar sequence to the amino acid sequence of the protein monomer is searched for in an analyzed protein structure database, and the distance between the residues is extracted based on a search method (HHSearch) of a hidden Markov model (HMM) using a tool for protein sequence analysis, and is used as a template feature.
[0048] In some embodiments, the amino acid sequence of each protein monomer is input into two pre-configured linear networks to obtain candidate sequence coding features. One empty dimension is added in different directions of the candidate sequence coding features to obtain the first sequence coding feature and the second sequence coding feature, and the first sequence coding feature and the second sequence coding feature are added to obtain the pair feature of each protein monomer. A plurality of sequences, a complex sequence with a length of r after splicing, is coded by two linear network Linear layers to obtain sequence coding features with a shape of [r,c], i.e., the first sequence coding feature z1 and the second sequence coding feature z2, where c is the hidden layer depth (hyper parameter) of the linear network, and then one empty dimension is added to z1 and z2 (the shape of z1 is converted to [r,1,c] and the shape of z2 is converted to [1,r,c]), and then added to obtain the pair feature z pair where z pair has shape [r,r,c], and zpair =z1+z2.
[0049] In S402, the template features of each protein monomer are input to the linear network for mapping, and then added to the pair features of each protein monomer to obtain candidate residue pair features.
[0050] In the embodiment of the present disclosure, the shape of the Template feature after splicing is [r, r], and after being encoded by the Linear layer, a feature z temp Obtaining the feature z temp and the pair feature to obtain the candidate residue pair feature.
[0051] In S403, the candidate residue pair features are input to a preset encoder for encoding to obtain a target residue pair feature of each protein monomer.
[0052] In an embodiment of the present disclosure, the candidate residue pair features are input to an encoder (Evofomer Encoder) and encoded to obtain the target residue pair features for each protein monomer.
[0053] In S404, based on the target amino acid sequence of each protein monomer, a homologous sequence of each protein monomer is searched and obtained from a plurality of gene sequence databases.
[0054] The present invention first uses the amino acid sequence of each protein monomer in the complex as a query request to search for homologous sequences in multiple gene sequence databases. By using existing tools JackHMMER and HHblits, deeper analysis and annotation of protein sequences can be achieved. JackHMMER can perform a fast heuristic search of hidden Markov models (HMMs), and HHblits can be used to perform more detailed annotation of the discovered protein sequences, thereby obtaining the homologous sequences of each protein monomer.
[0055] In S405, multiple sequence alignment is performed for the homologous sequences of each protein monomer to obtain candidate multiple sequence alignment features of each protein monomer.
[0056] In the resulting homologous sequences, multiple sequence alignment is used to obtain a multiple sequence alignment feature (MSA) for each monomer.
[0057] In S406, the candidate multiple sequence alignment features of each protein monomer are input to a pre-configured encoder and encoded to obtain a target multiple sequence alignment feature of each protein monomer.
[0058] The candidate multiple sequence alignment features of each protein monomer are input to an encoder (Evofomer Encoder) and encoded to obtain the target multiple sequence alignment features of each protein monomer.
[0059] S407, regularizing the target multiple sequence alignment feature of each protein monomer to obtain a first multiple sequence alignment feature of each protein monomer, and mapping the target multiple sequence alignment feature of each protein monomer to obtain a second multiple sequence alignment feature of each protein monomer.
[0060] A regularization Norm process is performed on the target multiple sequence alignment feature of each protein monomer to obtain a first multiple sequence alignment feature of each protein monomer, and the target multiple sequence alignment feature of each protein monomer is mapped based on a linear network to obtain a second multiple sequence alignment feature of each protein monomer.
[0061] In S408, the initial coordinates of each amino acid residue, the target residue pair features of each protein monomer, the first multiple sequence alignment feature and the second multiple sequence alignment feature are input into an N-stage folded repeat network layer, and the torsion angle, residue-level position transformation and monomer chain-level position transformation of each amino acid residue are predicted by the N-stage folded repeat network layer to obtain the target coordinates of each amino acid residue, and a predicted structure of the protein complex is obtained.
[0062] For an explanation of step S408, please refer to the relevant content in the above embodiment, and a detailed explanation will be omitted here.
[0063] In the embodiments of the present application, the initial coordinates of each amino acid residue in a target protein complex are obtained, and the target residue pair features, first multiple sequence alignment features and second multiple sequence alignment features of each protein monomer in the target protein complex are obtained, which can effectively promote the development of the protein monomer structure prediction task and improve the overall effect of protein structure prediction.
[0064] FIG. 5 is a structural diagram of a method for predicting a protein complex structure according to an embodiment of the present disclosure. As shown in FIG. 5, in the embodiment of the present disclosure, based on the amino acid sequence [Sequence 1, ..., Sequence N] of each of the protein monomers in the target protein complex, a homologous sequence is searched from a plurality of gene sequence databases (Sequence Data Base), and a multiple sequence alignment is performed to obtain a multiple sequence alignment feature [MSA 1, ..., MSA N] of each of the protein monomers. The multiple sequence alignment feature [MSA 1, ..., MSA N] is input to an Evofomer Encoder for encoding to obtain a target multiple sequence alignment feature. The amino acid sequence of each of the protein monomers is input to two pre-defined linear networks Linear to generate a pair feature. For the target amino acid sequence of each of the protein monomers, a protein structure having a similar sequence is searched from a protein structure database (Structure Data Base), and the distance between residues is extracted as a template feature. Optionally, Pair and Merge indicate merging, that is, MSA representation feature, pair It shows that a merging process is performed on the representation features, and the template features of each of the protein monomers are input to a linear network for mapping, and then added to the pair features of each of the protein monomers, and then input to an Evofomer encoder for encoding to obtain the target residue pair features of each of the protein monomers. The Evofomer encoder can extract hidden layer codes for each residue from MSA, Pair and Template data, and in the decoding stage, the present disclosure can support the training task by MSA masking prediction (Mask MSA), LDDT prediction (LDDT is a measurement method that exists in advance and is used in the field of protein structure prediction), residue distance prediction, etc. based on the structure prediction module (Structure Module) in the protein complex structure prediction AF2 Multimer model. Here, the initial frame represents the initial coordinates.
[0065] As shown in Figure 5, a protein complex containing r residues is input, and after being processed by Evofomer Encoder, the model obtains the MSA feature code of each residue i of the target protein complex, i.e., the target multiple sequence alignment feature
[0066]
number
[0067]
number
[0068] In an embodiment of the present disclosure, an N-stage folding iteration network layer (Fold Iteration module) is used to predict the structure of a target protein complex, and optionally, the value of N may be 8, although in other implementations, N may be other values and is not limited in the embodiments of the present disclosure.
[0069] To accommodate the rotational invariance of protein structures, the present invention uses a relative position transformation T i =(R i ,t(→) i ) is used to represent the coordinates of each residue, and the coordinate origin T i= (I,0(→)) to initialize the spatial structure of the protein complex. The present invention first uses two regularized norm layer networks
[0070]
number
[0071]
number
[0072] Here, each fold iteration is s i , Z i,j , T i After accessing the sigma-based sigma-based matrix, we first extract the rotationally invariant residue code s by the invariant point attention module. i After passing through network layers such as the Linear layer, Norm layer, and Dropout layer, the present disclosure uses the obtained code to obtain the torsion angle a f i ∈ R 2 and predict each residue coordinate TC k = (RC k, tC k), where the Dropout layer is used to randomly discard network parameters and plays a small role, so it is not shown in Figure 5 and this layer may be omitted or removed.
[0073] As shown in FIG. 5, in the embodiment of the present disclosure, in each folding iteration, a torsion angle a f i ∈ R in the side chain of residue i is predicted using a side chain and torsion angle predictor. 2 where
[0074]
number
[0075] In predicting the protein complex backbone network structure, the present disclosure first encodes the residue features using a shallow neural network (Linear and Norm) structure, and then uses the BackboneUpdate algorithm to estimate the Euclidean transformation T i , i.e., predict the first position transformation. In BackboneUpdate, we map the hidden layer features to a single 6-dimensional representation, where the first three dimensions b i , c i , d i is the rotation matrix R of residue i through the equation i and the last three dimensions t(→) i represents the residue translation transformation directly. Here, the input of Fold Iterationd does not consider the side chains and torsion angles predicted in the previous step, but only the main chain position transformation affine and residue code.
[0076] Prediction of spatial position transformation of residue i TR i = (R i ,t(→) i ), the model is i =T i O Based on TR i, we update the relative positions of the residues to complete the residue-level position transformation. In Eq. i is the updated T i indicates the second T i is the T before the update i Shows.
[0077] Based on the above transformation, the present disclosure introduces a monomer chain-level position transformation Chainaffine module to predict chain-level residue transformation, and the Chainaffine module aims to predict the overall transformation TC k of a protein complex chain k. The module structure can be realized in various forms, for example, as shown in FIG. 3, which is the realization method of one Chainaffine module. i Taking input, the Chainaffine module first calculates s based on the residue position code. i into different monomer chains and calculate the average of all residue representations of the same monomer, followed by chain-level hidden layer representation
[0078]
number
[0079] After this, the Chainaffine module uses a multi-layer neural network structure to
[0080]
number
[0081] As shown in Figure 5, when updating intrachain residue positions, every residue i in chain k shares a transformation TC k and gets a chain-level second position transformation TC i for each residue. Finally, the model is i =T i O Update the position of each residue in the complex based on TC i. In the formula, the first T i is the updated T i represents the second T i is the T before the update i Here, TC i represents T iIt acts on the front of the chain, translating the residues in the chain around the origin.
[0082] After obtaining the Euclidean transformation of the backbone network and the torsion angles of the side chains, in the embodiment of the present application, the residue update module and the Update frame module are used to perform the T i and a f i are converted into the three-dimensional coordinates of each residue to complete the tertiary structure prediction of the protein, where the Angles module predicts the side-chain torsion angles, and the Coordinates convert module receives the main-chain transformation and the side-chain torsion angles and outputs the converted spatial coordinates.
[0083] FIG. 6 is a structural diagram of a protein complex structure prediction device according to an embodiment of the present disclosure. As shown in FIG. 6, the protein complex structure prediction device 600 includes: an acquisition module 610 for acquiring initial coordinates of each amino acid residue in the target protein complex, and acquiring a target residue pair feature, a first multiple sequence alignment feature and a second multiple sequence alignment feature of each protein monomer in the target protein complex; a structure prediction module 620 for inputting the initial coordinates of each amino acid residue, the target residue pair feature of each protein monomer, the first multiple sequence alignment feature and the second multiple sequence alignment feature into an N-stage folded repeat network layer, and predicting the torsion angle, residue-level position transformation and monomer chain-level position transformation of each amino acid residue through the N-stage folded repeat network layer to obtain the target coordinates of each amino acid residue, and obtain a predicted structure of the protein complex; Here, the first multiple sequence alignment feature is a regularized multiple sequence alignment feature, the second multiple sequence alignment feature is a mapped multiple sequence alignment feature, and N is an integer greater than 1.
[0084] In some embodiments, the structure prediction module 620 further comprises: Input the initial coordinates, the target residue pair features and the second multiple sequence alignment features into a first-stage folded repeat network layer, and perform residue-level position transformation prediction and monomer chain-level position transformation prediction for each amino acid residue to obtain a target residue code 1 and a candidate position transformation 1 of the first-stage folded repeat network layer; For the m-th folded repeat network layer, the target residue pair feature, the target residue code m-1 of the m-1-th folded repeat network layer, and the candidate position transformation m-1 are input into the m-th folded repeat network layer to perform residue-level position transformation prediction and monomer-chain-level position transformation prediction for each amino acid residue, to obtain the target residue code m and the candidate position transformation m of the m-th folded repeat network layer, where the value of m is 2 to N; The Nth stage folded repeat network layer performs side chain and torsion angle prediction for the first multiple sequence alignment feature and the target residue code N of the Nth stage folded repeat network layer to obtain the torsion angle of the side chain of each amino acid residue, and obtains the target coordinate of each amino acid residue based on the torsion angle of the side chain of each amino acid residue and the candidate position transformation N of the Nth stage folded repeat network layer.
[0085] In some embodiments, the structure prediction module 620 further comprises: The first-stage folding iterative network layer performs an invariant attention mechanism and a mapping process on the initial coordinates, the target residue pair features and the second multiple sequence alignment features to obtain a target residue code 1; Predicting residue-level positional transformations for the target residue code 1 to obtain a first positional transformation 1 for each amino acid residue; predicting monomer-chain-level positional transformations for the target residue code 1 to obtain a second positional transformation 1 for each amino acid residue; A position update is performed based on the first position transformation 1, the second position transformation 1, and the initial coordinates to obtain a candidate position transformation 1 for the first stage folding network layer.
[0086] In some embodiments, the structure prediction module 620 further comprises: Perform an invariant attention mechanism and a mapping process on the candidate position transformation m-1 of the m-1-th folding iterative network layer input to the m-th folding network layer and the target residue pair feature to obtain a target residue code m; performing residue-level positional transformation predictions on the target residue code m to obtain a first positional transformation m for each amino acid residue; and performing monomer-chain-level positional transformation predictions on the target residue code m to obtain a second positional transformation m for each amino acid residue; Based on the first position transformation m and the second position transformation m, a candidate position transformation m for the mth stage folding network layer is obtained.
[0087] In some embodiments, the structure prediction module 620 further comprises: Based on the backbone updating algorithm, the target residue code of each amino acid residue is mapped to obtain the first position transformation of each amino acid residue.
[0088] In some embodiments, the structure prediction module 620 further comprises: for each amino acid residue, splitting two or more adjacent amino acid residues into different monomer chains based on the target residue code of the amino acid residue; For a target amino acid residue in any one of the monomer chains, an average value calculation is performed on the target residue code of the target amino acid residue to obtain a chain-level candidate residue code, and the candidate residue code is mapped based on a multi-layer neural network structure to obtain a second position transformation of each amino acid residue in the monomer chain.
[0089] In some embodiments, the multi-layer neural network structure comprises a three-layer linear network; The structure prediction module 620 further comprises: inputting and mapping the candidate residue code to a first linear network to obtain a first transformed representation; Mapping the first transformation representation to a second linear network to obtain a second transformation representation; The first and second transformed representations are input to a third linear network and mapped to obtain a second positional transformation of each amino acid residue in the monomer chain.
[0090] In some embodiments, the acquisition module 610 further comprises: Obtain a template feature for each protein monomer, and construct a pair feature of the amino acid sequence of each protein monomer; The template features of each protein monomer are input into the linear network for mapping, and then the pair features of each protein monomer are added to obtain candidate residue pair features; The candidate residue pair features are input to a pre-configured encoder and encoded to obtain a target residue pair feature for each protein monomer.
[0091] In some embodiments, the acquisition module 610 includes: Matching the target amino acid sequence of each protein monomer with a plurality of first amino acid sequences in a protein structure database to obtain a second amino acid sequence having a similarity greater than a predetermined threshold value; Distances between the coordinates of amino acid residues in the second amino acid sequence are extracted as template features for each protein monomer.
[0092] In some embodiments, the acquisition module 610 further comprises: The amino acid sequence of each protein monomer is input into two pre-defined linear networks to obtain candidate sequence-encoding features; Adding one empty dimension in different directions of the candidate sequence-encoded feature to obtain a first sequence-encoded feature and a second sequence-encoded feature; The first sequence-encoded feature and the second sequence-encoded feature are added to obtain a paired feature for each protein monomer.
[0093] In some embodiments, the acquisition module 610 further comprises: Searching and obtaining homologous sequences of each protein monomer from multiple gene sequence databases based on the target amino acid sequence of each protein monomer; Performing multiple sequence alignments against the homologous sequences of each protein monomer to obtain candidate multiple sequence alignment features for each protein monomer; The candidate multiple sequence alignment features of each protein monomer are input to a pre-configured encoder for encoding to obtain a target multiple sequence alignment feature of each protein monomer; The target multiple sequence alignment feature of each protein monomer is regularized to obtain a first multiple sequence alignment feature of each protein monomer, and the target multiple sequence alignment feature of each protein monomer is mapped to obtain a second multiple sequence alignment feature of each protein monomer.
[0094] The present disclosure takes into account the relative independence of each monomer chain in a protein complex, and adds monomer chain level position conversion to the residue level position conversion to update the coordinates of each amino acid residue, thereby accurately predicting the structure of the predicted protein, improving the efficiency of protein complex structure prediction, and making it better applicable to application scenarios in which a protein complex contains multiple chains.
[0095] According to an embodiment of the present disclosure, the present disclosure further provides an electronic device and a readable storage medium. According to an embodiment of the present disclosure, the present disclosure provides a computer program, which, when executed by a processor, realizes the method for predicting a protein complex structure provided by the present disclosure.
[0096] 7 is a block diagram of an electronic device for implementing the embodiment of the present disclosure. The electronic device can implement the protein complex structure prediction method of the embodiment of the present disclosure, and the electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processors, mobile phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or claimed herein.
[0097] 7, the device 700 includes a computing unit 701 that can perform various suitable operations and processes according to a computer program stored in a read-only memory (ROM) 702 or loaded from a storage unit 708 into a random access memory (RAM) 703. The RAM 703 may also store various programs and data necessary for the operation of the device 700. The computing unit 701, the ROM 702 and the RAM 703 are connected to each other via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.
[0098] The multiple components of the device 700 are connected to an I / O interface 705, including an input unit 706 such as a keyboard, a mouse, etc., an output unit 707 such as various types of displays, speakers, etc., a storage unit 708 such as a magnetic disk, an optical disk, etc., and a communication unit 709 such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 709 enables the device 700 to exchange information / data with other equipment via a computer network such as the Internet and / or various telecommunication networks.
[0099] The computing unit 701 may be any of a variety of general-purpose and / or special-purpose processing components having processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, computing units that execute various machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 may perform the methods and processes described above, such as the text correction method. For example, in some embodiments, the text correction method may be embodied as a computer software program tangibly embodied in a machine-readable medium, such as the storage unit 708. In some embodiments, some or all of the computer program may be loaded and / or installed in the device 700 via the ROM 702 and / or the communication unit 709. When the computer program is loaded into the RAM 703 and executed by the computing unit 701, one or more steps of the text correction method described above may be performed. Optionally, in other embodiments, the computing unit 701 may be configured to perform the text correction method in any other suitable manner (eg, via firmware).
[0100] Various embodiments of the systems and techniques described herein may be implemented in digital electronic circuitry systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), load programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include being embodied in one or more computer programs that can be executed and / or interpreted on a programmable system that includes at least one programmable processor, which may be a special purpose or general purpose programmable processor, and that receives data and instructions from and transmits data and instructions to a storage system, at least one input device, and at least one output device.
[0101] The program codes for implementing the methods of the present disclosure can be programmed in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general purpose computer, a special purpose computer, or other programmable data processing apparatus, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are performed. The program codes may be executed entirely on the machine, partially on the machine, as a stand-alone software package, partially on the machine and partially on a remote machine, or entirely on a remote machine or server.
[0102] In the context of this disclosure, a machine-readable medium is a tangible medium that can contain or store a program for use by or in combination with an instruction execution system, device, or apparatus. A machine-readable medium is a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium includes, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or apparatus, or any suitable combination of the above. More specific examples of machine-readable storage media include, but are not limited to, an electrical connection based on one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above.
[0103] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to a user, and a keyboard and pointing device (e.g., a mouse or trackball) by which a user can provide input to the computer. Other types of devices can also provide interaction with a user, for example, feedback provided to the user can be any form of sensing feedback (e.g., vision feedback, auditory feedback, or haptic feedback), and input from the user can be received in any form (including sound input, speech input, or tactile input).
[0104] The systems and techniques described herein may be implemented in a computing system including a back-end component (e.g., as a data server), or a computing system including a middleware component (e.g., an application server), or a computing system including a front-end component (e.g., a user computer having a graphical user interface or a web browser through which a user interacts with an embodiment of the systems and techniques described herein), or a computing system including any combination of such back-end, middleware, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include a local area network (LAN), a wide area network (WAN), and the Internet.
[0105] The computer system may include a client and a server. The client and server are generally remote from each other and typically interact through a communication network. The relationship between client and server is created by computer programs running on corresponding computers and having a client-server relationship to each other. The server may be a cloud server, a server of a distributed system, or a server combined with a blockchain.
[0106] It should be noted that, using the various forms of flows shown above, steps can be rearranged, added, or deleted. For example, each step described in the present disclosure can be performed in parallel, sequentially, or in a different order, but is not limited herein as long as the desired results of the technical solutions disclosed in the present application can be achieved.
[0107] The above specific embodiments do not limit the scope of protection of the present disclosure. Those skilled in the art can make various modifications, combinations, subcombinations, and substitutions based on design requirements and other factors. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principle of the present disclosure should be included within the scope of protection of the present disclosure.
Claims
1. A method for predicting a protein complex structure, comprising: obtaining initial coordinates of each amino acid residue in a target protein complex; and obtaining a target residue pair feature, a first multiple sequence alignment feature, and a second multiple sequence alignment feature of each protein monomer in the target protein complex; inputting the initial coordinates of each of the amino acid residues, the target residue pair feature of each of the protein monomers, the first multiple sequence alignment feature and the second multiple sequence alignment feature into an N-stage folded repeat network layer; predicting the torsion angle, residue-level position transformation and monomer-chain-level position transformation of each of the amino acid residues by the N-stage folded repeat network layer to obtain the target coordinates of each of the amino acid residues, and obtaining a predicted structure of the protein complex; the first multiple sequence alignment feature is a regularized multiple sequence alignment feature, the second multiple sequence alignment feature is a mapped multiple sequence alignment feature, and N is an integer greater than 1; Methods for predicting protein complex structures.
2. inputting the initial coordinates, the target residue pair features and the second multiple sequence alignment features into a first-stage folded repeat network layer, and performing residue-level positional transformation prediction and monomer-chain-level positional transformation prediction for each of the amino acid residues to obtain a target residue code 1 and a candidate positional transformation 1 of the first-stage folded repeat network layer; For a folded repeat network layer of the mth stage, the target residue pair feature, the target residue code m-1 of the folded repeat network layer of the m-1th stage, and the candidate position transformation m-1 are input to the folded repeat network layer of the mth stage, and a residue level position transformation prediction and a monomer chain level position transformation prediction are performed for each amino acid residue to obtain a target residue code m and a candidate position transformation m of the folded repeat network layer of the mth stage, where the value of m is 2 to N; and performing side chain and torsion angle prediction for the first multiple sequence alignment feature and the target residue code N of the folded repeat network layer of the Nth stage by the folded repeat network layer of the Nth stage to obtain a torsion angle of the side chain of each of the amino acid residues, and obtaining a target coordinate of each of the amino acid residues based on the torsion angle of the side chain of each of the amino acid residues and the candidate position transformation N of the folded repeat network layer of the Nth stage. The method for predicting the structure of a protein complex according to claim 1.
3. inputting the initial coordinates, the target residue pair features and the second multiple sequence alignment features into a first-stage folded repeat network layer, and predicting residue-level positional transformations and monomer-chain-level positional transformations for each of the amino acid residues to obtain a target residue code 1 and a candidate positional transformation 1 of the first-stage folded repeat network layer; performing an invariant attention mechanism and a mapping process on the initial coordinates, the target residue pair features and the second multiple sequence alignment features by the first-stage folding iterative network layer to obtain the target residue code 1; performing residue-level positional transformation predictions on the target residue code 1 to obtain a first positional transformation 1 for each of the amino acid residues; and performing monomer-chain-level positional transformation predictions on the target residue code 1 to obtain a second positional transformation 1 for each of the amino acid residues; Performing a position update based on the first position transformation 1, the second position transformation 1, and the initial coordinate to obtain a candidate position transformation 1 of the first stage folded network layer; The method for predicting a protein complex structure according to claim 2.
4. inputting the target residue pair feature, the target residue code m-1 of the folded repeat network layer of the m-1 stage, and the candidate position transformation m-1 of the folded repeat network layer of the m stage, and predicting a residue level position transformation and a monomer chain level position transformation for each amino acid residue to obtain a target residue code m and a candidate position transformation m of the folded repeat network layer of the m stage; performing an invariant attention mechanism and a mapping process on the candidate position transformation m-1 of the (m-1)th folding iterative network layer input to the (m)th folding network layer and the target residue pair feature to obtain the target residue code m; performing residue-level positional transformation prediction for the target residue code m to obtain a first positional transformation m for each of the amino acid residues; and performing monomer-chain-level positional transformation prediction for the target residue code m to obtain a second positional transformation m for each of the amino acid residues; and obtaining a candidate position transformation m of the mth stage folded network layer based on the first position transformation m and the second position transformation m. The method for predicting a protein complex structure according to claim 2.
5. A process for predicting a residue-level positional transformation for a target residue code of each of the amino acid residues to obtain a first positional transformation for each of the amino acid residues includes: mapping a target residue code for each of the amino acid residues based on a backbone update algorithm to obtain the first position transformation for each of the amino acid residues; The method for predicting a protein complex structure according to claim 3.
6. The process of predicting a monomer-chain-level positional transformation for a target residue code of each of the amino acid residues to obtain a second positional transformation for each of the amino acid residues includes: for each said amino acid residue, splitting two or more adjacent said amino acid residues into different monomer chains based on the target residue code of said amino acid residue; For a target amino acid residue in any one of the monomer chains, performing an average calculation on the target residue code of the target amino acid residue to obtain a chain-level candidate residue code, and mapping the candidate residue code based on a multi-layer neural network structure to obtain the second position transformation of each of the amino acid residues in the monomer chain. The method for predicting a protein complex structure according to claim 3.
7. the multi-layer neural network structure comprises a three-layer linear network; Mapping the candidate residue codes based on the multi-layer neural network structure to obtain the second position transformation of each of the amino acid residues in the monomer chain, comprising: mapping the candidate residue codes to a first linear network to obtain a first transformed representation; mapping the first transformed representation through a second linear network to obtain a second transformed representation; and mapping the first transformed representation and the second transformed representation into a third linear network to obtain the second position transformation of each of the amino acid residues in the monomer chain. The method for predicting the structure of a protein complex according to claim 6.
8. The step of obtaining a target residue pair feature for each of the protein monomers in the target protein complex comprises: obtaining a template feature for each of the protein monomers and constructing a pair feature of the amino acid sequence of each of the protein monomers; inputting the template feature of each of the protein monomers into a linear network for mapping, and then adding it to the pair feature of each of the protein monomers to obtain a candidate residue pair feature; and inputting the candidate residue pair features into a pre-configured encoder to encode the candidate residue pair features to obtain a target residue pair feature for each of the protein monomers. The method for predicting the structure of a protein complex according to claim 1.
9. Obtaining a template feature for each of the protein monomers includes: Matching the target amino acid sequence of each of the protein monomers with a plurality of first amino acid sequences in a protein structure database to obtain second amino acid sequences having a similarity greater than a preset threshold value; extracting distances between coordinates of amino acid residues of the second amino acid sequence as template features for each of the protein monomers; The method for predicting a protein complex structure according to claim 8.
10. Constructing a pairwise signature of an amino acid sequence of each of said protein monomers comprises: inputting the amino acid sequence of each of the protein monomers into two pre-defined linear networks to obtain candidate sequence-encoding features; adding an empty dimension in each of different directions of the candidate sequence-encoded feature to obtain a first sequence-encoded feature and a second sequence-encoded feature; and adding the first sequence-encoding feature and the second sequence-encoding feature to obtain a paired feature for each of the protein monomers. The method for predicting a protein complex structure according to claim 8.
11. The step of obtaining a first multiple sequence alignment feature and a second multiple sequence alignment feature of each of the protein monomers in the target protein complex includes: searching and acquiring homologous sequences of each of the protein monomers from a plurality of gene sequence databases based on the target amino acid sequence of each of the protein monomers; performing multiple sequence alignment for each of said protein monomers against homologous sequences to obtain candidate multiple sequence alignment features for each of said protein monomers; inputting the candidate multiple sequence alignment features of each of the protein monomers into a preset encoder and encoding them to obtain a target multiple sequence alignment feature of each of the protein monomers; regularizing the target multiple sequence alignment feature of each of the protein monomers to obtain a first multiple sequence alignment feature of each of the protein monomers; and mapping the target multiple sequence alignment feature of each of the protein monomers to obtain a second multiple sequence alignment feature of each of the protein monomers. The method for predicting the structure of a protein complex according to claim 1.
12. A protein complex structure prediction device, comprising: an acquisition module for acquiring initial coordinates of each amino acid residue in a target protein complex, and acquiring a target residue pair feature, a first multiple sequence alignment feature, and a second multiple sequence alignment feature of each protein monomer in the target protein complex; a structure prediction module for inputting the initial coordinates of each of the amino acid residues, the target residue pair feature of each of the protein monomers, the first multiple sequence alignment feature and the second multiple sequence alignment feature into an N-stage folded repeat network layer, and predicting the torsion angle, residue-level position transformation and monomer chain-level position transformation of each of the amino acid residues through the N-stage folded repeat network layer to obtain the target coordinates of each of the amino acid residues, and obtaining a predicted structure of the protein complex; the first multiple sequence alignment feature is a regularized multiple sequence alignment feature, the second multiple sequence alignment feature is a mapped multiple sequence alignment feature, and N is an integer greater than 1; A protein complex structure prediction device.
13. The structure prediction module further comprises: inputting the initial coordinates, the target residue pair features and the second multiple sequence alignment features into a first-stage folded repeat network layer, and performing residue-level position transformation prediction and monomer-chain-level position transformation prediction for each of the amino acid residues to obtain a target residue code 1 and a candidate position transformation 1 of the first-stage folded repeat network layer; For a folded repeat network layer of the mth stage, the target residue pair feature, the target residue code m-1 of the folded repeat network layer of the m-1th stage, and the candidate position transformation m-1 are input to the folded repeat network layer of the mth stage, and a residue level position transformation prediction and a monomer chain level position transformation prediction are performed for each amino acid residue to obtain a target residue code m and a candidate position transformation m of the folded repeat network layer of the mth stage, the value of m being 2 to N; By the folded repeat network layer of the Nth stage, perform side chain and torsion angle prediction for the first multiple sequence alignment feature and the target residue code N of the folded repeat network layer of the Nth stage to obtain the torsion angle of the side chain of each of the amino acid residues, and obtain the target coordinate of each of the amino acid residues according to the torsion angle of the side chain of each of the amino acid residues and the candidate position transformation N of the folded repeat network layer of the Nth stage. The apparatus for predicting a protein complex structure according to claim 12.
14. The structure prediction module further comprises: The first-stage folding iterative network layer performs an invariant attention mechanism and a mapping process on the initial coordinates, the target residue pair features, and the second multiple sequence alignment features to obtain the target residue code 1; performing a residue-level positional transformation prediction for the target residue code 1 to obtain a first positional transformation 1 for each of the amino acid residues; performing a monomer-chain-level positional transformation prediction for the target residue code 1 to obtain a second positional transformation 1 for each of the amino acid residues; Perform location update based on the first location transformation 1, the second location transformation 1, and the initial coordinates to obtain a candidate location transformation 1 of the first stage folded network layer; The apparatus for predicting a protein complex structure according to claim 13.
15. The structure prediction module comprises: Performing an invariant attention mechanism and a mapping process on the candidate position transformation m-1 of the m-1th folding iterative network layer input to the mth folding network layer and the target residue pair feature to obtain the target residue code m; performing a residue-level positional transformation prediction for the target residue code m to obtain a first positional transformation m for each of the amino acid residues; and performing a monomer-chain-level positional transformation prediction for the target residue code m to obtain a second positional transformation m for each of the amino acid residues; Obtain a candidate position transformation m of the mth stage folded network layer based on the first position transformation m and the second position transformation m. The apparatus for predicting a protein complex structure according to claim 13.
16. The structure prediction module further comprises: mapping a target residue code for each of the amino acid residues based on a backbone update algorithm to obtain the first position transformation for each of the amino acid residues; The apparatus for predicting a protein complex structure according to claim 14.
17. The structure prediction module further comprises: for each said amino acid residue, splitting two or more adjacent said amino acid residues onto different monomer chains based on the target residue code of said amino acid residue; For a target amino acid residue in any one of the monomer chains, perform an average calculation on the target residue code of the target amino acid residue to obtain a chain-level candidate residue code, and map the candidate residue code based on a multi-layer neural network structure to obtain the second position transformation of each of the amino acid residues in the monomer chain. The apparatus for predicting a protein complex structure according to claim 14.
18. the multi-layer neural network structure comprises a three-layer linear network; The structure prediction module further comprises: mapping the candidate residue code to a first linear network to obtain a first transformed representation; Mapping the first transformed representation through a second linear network to obtain a second transformed representation; mapping the first transformed representation and the second transformed representation into a third linear network to obtain the second position transformation of each of the amino acid residues in the monomer chain; The apparatus for predicting a protein complex structure according to claim 17.
19. The acquisition module further comprises: Obtaining a template feature for each of the protein monomers and constructing a pair feature of an amino acid sequence for each of the protein monomers; inputting the template feature of each of the protein monomers into a linear network for mapping, and then adding it to the pair feature of each of the protein monomers to obtain a candidate residue pair feature; inputting the candidate residue pair features into a pre-configured encoder to encode the candidate residue pair features to obtain a target residue pair feature for each of the protein monomers; The apparatus for predicting a protein complex structure according to claim 12.
20. The acquisition module: Matching the target amino acid sequence of each of the protein monomers with a plurality of first amino acid sequences in a protein structure database to obtain a second amino acid sequence having a similarity greater than a predetermined threshold value; extracting distances between coordinates of amino acid residues of the second amino acid sequence as template features for each of the protein monomers; The apparatus for predicting a protein complex structure according to claim 19.
21. The acquisition module further comprises: Inputting the amino acid sequence of each of the protein monomers into two pre-designed linear networks to obtain candidate sequence-encoding features; adding an empty dimension in each of different directions of the candidate sequence-encoded feature to obtain a first sequence-encoded feature and a second sequence-encoded feature; adding the first sequence-encoding feature and the second sequence-encoding feature to obtain a paired feature for each of the protein monomers. The apparatus for predicting a protein complex structure according to claim 19.
22. The acquisition module further comprises: Searching and obtaining homologous sequences for each of the protein monomers from a plurality of gene sequence databases based on the target amino acid sequence of each of the protein monomers; performing a multiple sequence alignment for each of the homologous sequences of the protein monomers to obtain a candidate multiple sequence alignment feature for each of the protein monomers; The candidate multiple sequence alignment features of each of the protein monomers are input to a preset encoder and encoded to obtain a target multiple sequence alignment feature of each of the protein monomers; regularizing the target multiple sequence alignment feature of each of said protein monomers to obtain a first multiple sequence alignment feature of each of said protein monomers; and mapping the target multiple sequence alignment feature of each of said protein monomers to obtain a second multiple sequence alignment feature of each of said protein monomers. The apparatus for predicting a protein complex structure according to claim 12.
23. 1. An electronic device comprising: At least one processor; a memory communicatively coupled to the at least one processor; The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to cause the at least one processor to perform a method according to any one of claims 1 to 11. Electronic devices.
24. A non-transitory computer-readable storage medium having computer instructions stored thereon, comprising: The computer instructions cause a computer to carry out a method according to any one of claims 1 to 11. A non-transitory computer-readable storage medium.
25. A computer program comprising: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 11 are realized. Computer program.
Citation Information
Patent Citations
Method for predicting distance between protein residues based on self-attention mechanism
CN114708903A
Protein complex structure prediction method
CN116779038A
Predicting protein structures over multiple iterations using recycling
WO2022112260A1
