Protein interface residue prediction method and device, electronic equipment and storage medium

Abstract

The invention provides an interface residue prediction method and device of protein, electronic equipment and a storage medium. The method comprises the following steps: extracting an initial coding feature of each amino acid residue in to-be-predicted protein chain data; constructing a local environment structure diagram of each amino acid residue by taking each amino acid residue as a central node and taking adjacent residues of the central node as connecting nodes; based on the initial coding feature of the central node and the initial coding feature of each connection node, performing fusion to obtain a fusion residue feature of the central node; and determining an interface residue prediction result of each amino acid residue based on the fused residue characteristics. According to the method provided by the invention, the contextual information of the central node is learned to obtain the fused residue characteristics of the central node, and the interface residue prediction result of each amino acid residue is determined based on the fused residue characteristics, so that the characteristics of each adjacent amino acid residue in an amino acid residue body and a local environment structure chart are fully utilized; and the prediction accuracy of the interface fusion residues is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of bioinformatics computing technology, and in particular to a method, device, electronic device and storage medium for predicting interface residues of a protein. Background Art

[0002] Accurately predicting the residues on proteins that interact with other proteins, nucleic acids, and ligands is of great significance for understanding how proteins function in cells, proteomics research, and drug design. Machine learning methods have been widely used to predict protein interface residues, and usually use artificially customized features directly to represent the sequence and structural information of proteins. For example, GraphPPIS selects features such as PSSM (Position-Specific Scoring Matrix), HMM (Hidden Markov Model), secondary structure, main chain torsion angle, and relative accessible surface area to represent the sequence and structural information of proteins.

[0003] However, artificially selected features are somewhat arbitrary. Only artificially customized features may not fully utilize the useful information in the protein sequence and structure for downstream task predictions, which in turn leads to the need to improve the accuracy of protein interface residue predictions. Summary of the invention

[0004] The present invention provides a method, device, electronic device and storage medium for predicting the interface residues of a protein, so as to solve the defect of low accuracy in predicting the interface residues of a protein in the prior art.

[0005] The present invention provides a method for predicting interface residues of a protein, comprising: Acquiring the protein chain data to be predicted, and extracting the initial coding features of each amino acid residue in the protein chain data to be predicted; Taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes, a local environment structure diagram of each amino acid residue is constructed; Based on the initial coding feature of the central node and the initial coding features of each connecting node, a fusion residue feature of the central node is obtained; Based on the fusion residue features, the interface residue prediction results of each amino acid residue in the protein chain data to be predicted are determined.

[0006] According to a protein interface residue prediction method provided by the present invention, the initial coding feature based on the central node is fused with the initial coding features of each connecting node to obtain the fused residue feature of the central node, including: Determining attention weights of the central node and the connecting nodes based on the initial coding features of the central node and the initial coding features of the connecting nodes; Based on the attention weight, the initial coding feature of the central node and the initial coding features of each connection node are weightedly fused to obtain the fused residue feature.

[0007] According to a protein interface residue prediction method provided by the present invention, the initial coding feature of the central node and the initial coding features of each connecting node are weightedly fused based on the attention weight to obtain the fused residue feature, including: Based on the graph neural network and the attention weight, the initial coding features of the central node and the initial coding features of each connection node are weightedly fused to obtain an initial fused feature; The initial fusion features are average-pooled to obtain the fusion residue features.

[0008] According to a method for predicting interface residues of a protein provided by the present invention, the protein chain data to be predicted includes: a protein chain sequence to be predicted and a protein chain structure to be predicted; The extracting and obtaining the initial coding features of each amino acid residue in the protein chain data to be predicted includes: Based on the coding model, the sequence characteristics and structural characteristics of each amino acid residue in the protein chain data to be predicted are extracted; The sequence features and structural features of the amino acid residues are spliced ​​to obtain the initial coding features.

[0009] According to a method for predicting interface residues of a protein provided by the present invention, the method of determining the interface residue prediction results of each amino acid residue in the protein chain data to be predicted based on the fusion residue features comprises: Inputting the fusion residue features into an interface residue prediction model, and outputting interface residue prediction results of each amino acid residue based on the interface residue prediction model; The interface residue prediction result is a protein interface residue prediction result, a DNA interface residue prediction result, or an RNA interface residue prediction result; The interface residue prediction model is trained based on sample protein complex data.

[0010] According to a protein interface residue prediction method provided by the present invention, the sample protein complex data includes at least one of first protein complex data, second protein complex data and third protein complex data; the first protein complex data is obtained based on protein-protein construction; the second protein complex data is obtained based on protein-DNA construction; the third protein complex data is obtained based on protein-RNA construction; The step of acquiring the sample protein complex data comprises: Obtain initial sample protein complex data; The sample protein complex data are obtained by screening the initial sample protein complex data based on the sequence consistency between the initial sample protein complex data.

[0011] The present invention also provides a protein interface residue prediction device, comprising: An acquisition unit, which acquires the protein chain data to be predicted and extracts the initial coding features of each amino acid residue in the protein chain data to be predicted; A graph construction unit, taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes, to construct a local environment structure graph of each amino acid residue; A fusion unit, which fuses the initial coding features of the central node with the initial coding features of each connecting node to obtain a fusion residue feature of the central node; The prediction unit determines the interface residue prediction result of each amino acid residue in the protein chain data to be predicted based on the fusion residue feature.

[0012] The present invention also provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the method for predicting interface residues of a protein as described above is implemented.

[0013] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the above-described methods for predicting interface residues of a protein.

[0014] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the method for predicting interface residues of a protein as described above is implemented.

[0015] The protein interface residue prediction method, device, electronic device and storage medium provided by the present invention construct a local environment structure diagram of each amino acid residue by taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes, obtain the fusion residue characteristics of the central node by learning the context information of the central node, and determine the interface residue prediction result of each amino acid residue based on the fusion residue characteristics, making full use of the characteristics of the amino acid residue itself and each adjacent amino acid residue in the local environment structure diagram, greatly improving the accuracy of interface fusion residue prediction. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0017] Figure 1 This is one of the schematic diagrams of the process of predicting the interface residues of a protein provided by the present invention; Figure 2 This is the second schematic diagram of the process of predicting the interface residues of a protein provided by the present invention; Figure 3 It is a schematic diagram of the structure of the interface residue prediction model provided by the present invention; Figure 4 It is a schematic diagram of the structure of the protein interface residue prediction device provided by the present invention; Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0018] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0019] Traditionally, in order to reliably determine protein interface residues, high-resolution structures of protein complexes need to be resolved by X-ray crystallography. However, resolving protein complexes is an expensive and time-consuming process, and the resolution of some transient complexes is particularly difficult. With the advancement of high-throughput experimental methods, such as ChIP-seq, CLIP-seq, and two-hybrid experiments, a large number of protein-DNA, protein-RNA, and protein-protein interaction pairs have been identified, but the protein interface residues involved in these interactions remain to be identified.

[0020] Currently, computational methods for predicting protein interface residues can be divided into two categories: sequence-based methods and structure-based methods. Among them, sequence-based methods are suitable for predicting interface residues of proteins with unknown three-dimensional structures, and their number is more than three orders of magnitude higher than that of known three-dimensional structures. For sequence-based methods, some sequence-derived features, such as evolutionary information, predicted secondary structure, solvent accessibility, and physicochemical properties, are usually used to predict protein interface residues. In general, sequence-based methods perform worse than structure-based methods because of the lack of three-dimensional structural information of proteins, and the structural features of protein surfaces play a key role in proteins recognizing specific binding partners. In addition, for structure-based methods, in addition to sequence-derived features, structure-derived features such as secondary structure, backbone torsion angles, and relative accessible surface area are also used.

[0021] However, whether it is a sequence-based method or a structure-based method, most of them require artificial definition of the displayed sequence and structural features, which to a certain extent limits the expressive power of the model, and thus the accuracy of downstream prediction tasks based on the extracted sequence and structural features needs to be improved.

[0022] In view of the above problems, the present invention provides a method for predicting interface residues of proteins to achieve accurate prediction of interface residues. Figure 1 FIG. 1 is one of the flow charts of the method for predicting the interface residues of a protein provided by the present invention, such as Figure 1 As shown, the method includes: Step 110, obtaining the protein chain data to be predicted, and extracting the initial coding features of each amino acid residue in the protein chain data to be predicted; Here, the protein chain data to be predicted may include the protein chain sequence and protein chain structure to be predicted. Among them, the protein chain sequence may refer to the order of clearance of different types of amino acids on the protein chain. For example, the protein chain sequence may be "GAMGLLDRT...". The protein chain structure may be the spatial position of the entire protein chain in three-dimensional space. The protein chain data to be predicted may be a currently existing protein chain or a designed protein chain, such as a protein chain predicted based on AlphaFold2.

[0023] In addition, the initial coding features here refer to the sequence features and structural features of each amino acid residue in the protein chain, which can be regarded as the coding features obtained by preliminary extraction through the encoder. It can be understood that amino acids remove a molecule of water through condensation reaction to form a peptide bond (-CO-NH-), thereby connecting into a polypeptide chain, and each amino acid unit after the condensation reaction is called an amino acid residue.

[0024] Specifically, a protein chain can be obtained from an existing protein chain database as the protein chain data to be predicted, or a designed protein chain can be obtained as the protein chain data to be predicted. Then, the sequence features and structural features of each amino acid residue in the protein chain data to be predicted can be extracted through the coding model as the initial coding features of the amino acid residues.

[0025] It should be noted that the amino acid residues in the protein chain that interact with other molecules are interface residues, which can also be referred to as binding interface residues. Generally speaking, interface residues have amino acid residues with similar co-evolutionary histories and are close to each other in the three-dimensional structure. In addition, the interacting interface residues are usually located on the surface of the protein and usually have a specific shape so as to form a stable binding interface with the ligand. Thus, the initial coding features here can reflect the evolutionary information of each amino acid residue and the three-dimensional structural information of each amino acid residue, so as to predict whether each amino acid residue is an interface residue.

[0026] Step 120, constructing a local environment structure diagram of each amino acid residue by taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes; Here, the adjacent residues of the central node can be determined by the distance between the main chain atoms of the two residues. For example, when the distance between the main chain atom of the amino acid residue and the main chain atom of the central node is less than a preset threshold, the amino acid residue can be considered as the adjacent residue of the central node. The main chain atom here can be a Cα atom. The preset distance here can be determined according to the type of macromolecule bound to the protein. For example, the preset distances corresponding to protein-protein binding, protein-DNA binding, and protein-RNA binding can be 10, ... , 14 and 12 Thus, the adjacent residues to the central node here include not only the amino acid residues directly connected to the central node, but also the amino acid residues whose distances to the main chain atoms of the central node are less than a preset threshold.

[0027] Specifically, in order to better represent the local environment of amino acid residues, each amino acid residue in the protein chain can be used as the central node, the adjacent residues of the central node as the connecting nodes, and the peptide bonds between the central node and the adjacent residues as the connecting edges to construct the local environment structure diagram of each amino acid residue. Compared with a single central node, the local environment structure diagram can more fully and comprehensively express the sequence information and structural information of the central node, thereby improving the accuracy of the prediction task.

[0028] Step 130, fusing the initial coding feature of the central node with the initial coding features of each connecting node to obtain a fused residue feature of the central node; Specifically, the initial coding features of the central node and the initial coding features of each connecting node can be fused to obtain the fused residue features of the central node and each connecting node, so as to obtain sequence information and three-dimensional structure information that is more relevant to the central node.

[0029] It should be noted that, compared with the prior art which only uses the sequence information and structural information of a single node, the embodiment of the present invention constructs a local environment graph of the central node, and fuses the initial coding features of each connected node in the local environment graph with the initial coding features of the central node to enrich the sequence information and three-dimensional structural information of the central node, learns the contextual sequence information of the central node and the three-dimensional spatial information of the local environment, and greatly improves the accuracy of interface fusion residue prediction.

[0030] Step 140: determining the interface residue prediction result of each amino acid residue in the protein chain data to be predicted based on the fusion residue feature.

[0031] Specifically, the prediction of which residues on a protein are interface residues can be regarded as a binary classification problem. Then, the fusion residue features can be input into a pre-trained classification model, and the classification model can be used to predict whether the amino acid residue is an interface residue based on the fusion residue features of the amino acid residue, thereby obtaining an interface residue prediction result. For example, the fusion residue features can be input into a multi-layer perceptron (MLP) to predict the probability that the amino acid residue is an interface residue. The prediction probability here can be calculated using the following formula, as shown below: ; in, represents the predicted probability; represents the activation function; , represents learnable parameters; Indicates fusion residue features.

[0032] The method provided by the embodiment of the present invention constructs a local environment structure diagram of each amino acid residue by taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes, obtains the fusion residue characteristics of the central node by learning the context information of the central node, and determines the interface residue prediction result of each amino acid residue based on the fusion residue characteristics, thereby making full use of the characteristics of the amino acid residue itself and the adjacent amino acid residues in the local environment structure diagram, greatly improving the accuracy of interface fusion residue prediction.

[0033] Based on any of the above implementation examples, step 130 includes: Determining attention weights of the central node and the connecting nodes based on the initial coding features of the central node and the initial coding features of the connecting nodes; Based on the attention weight, the initial coding feature of the central node and the initial coding features of each connection node are weightedly fused to obtain the fused residue feature.

[0034] Specifically, the attention weight between the central node and each connected node can be calculated by comparing the initial coding features of the central node with the initial coding features of each connected node. The attention weight here can be used to reflect the correlation and importance between the central node and each connected node in the local environment structure graph, enhance the feature representation of the connected nodes related to the central node in the fusion residue feature, and thus improve the accuracy of interface residue prediction based on the fusion residue feature.

[0035] In one embodiment, the attention weight may be calculated by the following formula, as shown below: ; In the formula, Indicates The connection node and Attention weights between central nodes; , represents learnable parameters; Indicates The initial encoding features of the central nodes; Indicates The initial encoding features of the connected nodes; represents the scaling factor, It can be 16.

[0036] Next, the initial encoding features of the central node and each connecting node can be weighted and fused by the attention weight between the central node and each connecting node to obtain a fused residue feature. In one embodiment, the fused residue feature can be calculated by the following formula, as shown in the following formula: ; In the formula, Indicates The central node and The fusion residue features of the connection nodes; , represents learnable parameters; Indicates The total number of connected nodes in the local environment structure graph of the central node.

[0037] Based on any of the above embodiments, the step of weightedly fusing the initial coding features of the central node and the initial coding features of each connection node based on the attention weight to obtain the fused residue feature includes: Based on the graph neural network and the attention weight, the initial coding features of the central node and the initial coding features of each connection node are weightedly fused to obtain an initial fused feature; The initial fusion features are average-pooled to obtain the fusion residue features.

[0038] Specifically, the initial coding features of each node in the local environment structure graph can be processed by a graph neural network to obtain a fused residual feature. In detail, the initial coding features can be weighted fused by a graph convolutional network, that is, based on the attention weight between the central node and each connected node, the initial feature coding of the central node and each connected node is weighted fused to obtain the initial fused feature.

[0039] Then, after completing the graph convolution operation, the entire local environment structure graph can be averaged by pooling, that is, the initial fusion features obtained by weighted fusion of the central node and each connected node are averaged to obtain the fusion residue features of the central node. The average pooling operation here can be calculated by the following formula, as shown below: ; In the formula, represents the fusion residue features of the central node; Indicates The total number of connected nodes in the local environment structure graph of the central node; express The initial fusion features of the connected nodes and the central node.

[0040] Based on any of the above embodiments, the protein chain data to be predicted includes: the protein chain sequence to be predicted and the protein chain structure to be predicted; The extracting and obtaining the initial coding features of each amino acid residue in the protein chain data to be predicted includes: Based on the coding model, the sequence characteristics and structural characteristics of each amino acid residue in the protein chain data to be predicted are extracted; The sequence features and structural features of the amino acid residues are spliced ​​to obtain the initial coding features.

[0041] Specifically, the protein chain sequence to be predicted can be input into a pre-trained coding model, and the sequence features of each amino acid residue in the protein chain can be extracted through the coding model. The pre-trained coding model here can be ESM2 (Evolutionary Sequence Model 2). The ESM2 model is a bioinformatics model based on deep learning, which can be used for the analysis and prediction of protein sequences. Similarly, the protein chain structure to be predicted can be input into a pre-trained coding model, and the structural features of each amino acid residue in the protein chain can be extracted through the coding model. The pre-trained coding model here can be ABACUS-R, which is an algorithm based on deep learning and can be used for encoding the structural features of amino acid residues. Finally, the sequence features and structural features of each amino acid residue can be spliced ​​to obtain the initial coding features of the amino acid residues. The initial coding features here can be expressed by the following formula, as shown below: ; In the formula, represents the initial encoding features; Represents the encoding vector of ESM2, i.e., the sequence feature; Represents the encoding vector of ABACUS-R, i.e., the structural feature.

[0042] It should be noted that in the prior art, artificially customized features are usually used to represent the sequence information and structural information of proteins, and the customized features are arbitrary to a certain extent. However, the present invention uses a pre-trained encoding model to perform sequence encoding and structural encoding on the protein chain data to be predicted, thereby obtaining comprehensive and rich sequence features and structural features, avoiding the omission of useful sequence information and structural information in the protein chain, and further improving the accuracy of the interface residue prediction task based on sequence features and structural features.

[0043] It should also be noted that compared with features that rely on PSSM and HMM, which require homologous sequence search and alignment, the speed of extracting sequence features and structural features through pre-trained encoding models is faster, and it is more suitable for high-throughput screening on a proteomics scale, and can improve the execution efficiency of interface residue prediction tasks.

[0044] Based on any of the above embodiments, the step of determining the interface residue prediction result of each amino acid residue in the protein chain data to be predicted based on the fusion residue feature comprises: Inputting the fusion residue features into an interface residue prediction model, and outputting interface residue prediction results of each amino acid residue based on the interface residue prediction model; The interface residue prediction result is a protein interface residue prediction result, a DNA interface residue prediction result, or an RNA interface residue prediction result; The interface residue prediction model is trained based on sample protein complex data.

[0045] It should be noted that proteins can form protein-protein complexes with other proteins, proteins can also form protein-DNA complexes with DNA, or proteins can also form protein-RNA complexes with RNA. Therefore, in the training stage, the sample protein complex data corresponding to the protein complex can be used to train and obtain respective interface residue prediction models. For example, the interface residue prediction models for predicting the protein binding interface, DNA binding interface, and RNA binding interface can be trained and obtained respectively.

[0046] Specifically, when performing the interface residue prediction task, the corresponding interface residue prediction model is selected according to the type of protein complex to be predicted. Then, the protein chain data to be predicted can be input into the corresponding interface residue prediction model, feature extraction and fusion are performed through the interface residue prediction model, and interface residue prediction is performed based on the fusion residue features to obtain the interface residue prediction results of each amino acid residue. Therefore, the interface residue prediction results here can be protein interface residue prediction results, or DNA interface residue prediction results, or RNA interface residue prediction results. It should be noted that the interface residue prediction model can also complete the feature extraction and fusion of protein chain data.

[0047] In one embodiment, Figure 2 FIG. 2 is a flow chart of the method for predicting the interface residues of a protein provided by the present invention. Figure 2 As shown, first, the protein sequence to be predicted ("GAMGLDP...") and the protein structure to be predicted can be input into the interface residue prediction model, and feature extraction (Feature extraction) is performed through the interface residue prediction model. Finally, the interface residue prediction model is used to output the interface residue prediction results of each amino acid residue in the protein chain to be predicted. For example, a PDB file can be output, and the binding interface residues predicted by the interface residue prediction model can be visualized using the PyMOL software. For example, different colors can be used to represent the probability of each amino acid residue being an interface residue. Amino acid residues in the red and yellow areas indicate that they may be potential binding interface residues. The redder the color of the amino acid residue, the greater the probability that it is a binding interface residue.

[0048] in, Figure 3is a schematic diagram of the structure of the interface residue prediction model provided by the present invention, such as Figure 3 As shown, the model includes: feature extraction module, feature fusion module and prediction module.

[0049] Among them, the feature extraction module contains two pre-trained coding models, ESM2 and ABACUS-R, which are used to extract the sequence features and structural features of each amino acid residue in the protein chain, and splice the sequence features and structural features of each amino acid residue to obtain the initial coding features.

[0050] The feature fusion module can be constructed by a graph neural network, which is used to perform graph convolution on the initial coding features of each node in the constructed local environment structure graph, fuse the associated coding features, and obtain the fused residue features of each amino acid residue.

[0051] The prediction module can be constructed through a multi-layer perceptron network, which is used to classify based on the fusion residue features of each amino acid residue, and predict the probability of whether the amino acid residue is an interface residue amino acid, that is, to obtain an interface residue prediction result.

[0052] Specifically, the protein chain sequence to be predicted can be input into the ESM2 coding model, and the sequence features of each amino acid residue can be extracted by the ESM2 coding model. Similarly, the protein chain structure to be predicted is input into the ABACUS-R coding model, and the structural features of each amino acid residue are extracted by the ABACUS-R coding model. Then, the sequence features and structural features of each amino acid residue are spliced ​​to obtain the initial coding features of each amino acid residue. Further, each amino acid residue is taken as the central node, and the adjacent residues of the central node are taken as the connecting nodes to construct the local environment structure graph of each amino acid residue. Then, the constructed local environment structure graph can be input into the graph neural network to propagate the message between the central node and each connecting node to obtain the initial fusion feature of the central node. Then, the initial fusion feature is averaged and pooled to obtain the fusion residue feature of the central node. Finally, the fusion residue feature of each amino acid residue is input into the multi-layer perceptron (MLP) network to obtain the probability that the amino acid residue is an interface residue, and obtain the interface residue prediction result.

[0053] The method provided by the embodiment of the present invention performs sequence encoding and structural encoding on the protein chain data to be predicted through a pre-trained encoding model, and efficiently obtains comprehensive and rich sequence features and structural features. Then, by constructing a local environment structure graph of each amino acid residue, the features of the nodes in the local environment structure graph are propagated through a graph neural network to achieve feature fusion between nodes, and obtain richer and more comprehensive fusion residue features. In this way, efficient and rapid prediction of protein interface residues is achieved, and the accuracy of interface residue prediction based on fusion residue features is greatly improved.

[0054] Based on any of the above embodiments, the sample protein complex data includes at least one of first protein complex data, second protein complex data and third protein complex data; the first protein complex data is constructed based on proteins and proteins; the second protein complex data is constructed based on proteins and DNA; the third protein complex data is constructed based on proteins and RNA; The step of acquiring the sample protein complex data comprises: Obtain initial sample protein complex data; The sample protein complex data are obtained by screening the initial sample protein complex data based on the sequence consistency between the initial sample protein complex data.

[0055] Specifically, in the sample data preparation stage, the sample protein complex data may include at least one of the first protein complex data, the second protein complex data, and the third protein complex data; the first protein complex data is constructed based on protein and protein; the second protein complex data is constructed based on protein and DNA; the third protein complex data is constructed based on protein and RNA. That is, the sample protein complex data includes sample protein-protein complex data, sample protein-DNA complex data, and sample protein-RNA complex data. Therefore, the initial sample protein complex data here also includes three corresponding protein complex data.

[0056] When obtaining the initial sample protein complex data, the initial sample protein complex data of protein-protein can be obtained from the known protein-protein complexes in the PDB (Protein Data Bank) database. Similarly, the initial sample protein complex data of protein-DNA and protein-RNA can be obtained from the BioLip database.

[0057] Furthermore, based on the sequence consistency between the obtained initial sample protein chain data, the corresponding type of sample protein complex data can be screened from different types of initial sample protein complex data. For example, data in which the sequence consistency between proteins in the protein-protein initial sample protein complex data is less than 25% can be used as sample protein complex data. For example, three widely used benchmark data sets can be obtained: Dset_186, Dset_72, and Dset_164, which contain 186, 72, and 164 protein sequences, respectively.

[0058] Similarly, the data in which the sequence identity between proteins in the initial sample protein complex data of protein-DNA and protein-RNA is less than 30% can be used as sample protein-DNA complex data and sample protein-RNA complex data. For example, the DNA_Train_573, DNA_Test_129, RNA_Train_495, and RNA_Test_117 data sets in the GraphBind data set are obtained, which contain 573, 129, 495, and 117 protein sequences, respectively. After obtaining all the sample protein complex data, 350 protein sequences in these three types of data sets can be randomly selected as training sets, and the remaining 72 can be used as test sets.

[0059] Furthermore, the interface residues of the sample protein complex data can be annotated by performing X-ray crystallography analysis (resolution <3.0Å) on all sample protein complex data. For example, for the sample protein-protein complex data, for the residues on the surface of the complex protein (relative solvent accessibility <5%), if the absolute solvent accessibility loss is <1.0Ų, the residue is annotated as a protein binding interface residue. Similarly, for the sample protein-DNA complex data and the sample protein-RNA complex data, if the minimum atomic distance between a residue and nucleic acid is less than the sum of the van der Waals radii corresponding to the nucleic acid atoms plus 0.5Å, the residue will be regarded as a nucleic acid binding interface residue.

[0060] Based on any of the above embodiments, Figure 4 : is a schematic diagram of the structure of the protein interface residue prediction device provided by the present invention, such as Figure 4 As shown, the device comprises: An acquisition unit 410 acquires the protein chain data to be predicted, and extracts the initial coding features of each amino acid residue in the protein chain data to be predicted; A graph construction unit 420, which constructs a local environment structure graph of each amino acid residue by taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes; A fusion unit 430, fusing the initial coding feature of the central node with the initial coding features of each connection node to obtain a fused residue feature of the central node; The prediction unit 440 determines the interface residue prediction result of each amino acid residue in the protein chain data to be predicted based on the fusion residue feature.

[0061] The device provided by the embodiment of the present invention constructs a local environment structure diagram of each amino acid residue by taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes, obtains the fusion residue characteristics of the central node by learning the context information of the central node, and determines the interface residue prediction result of each amino acid residue based on the fusion residue characteristics, thereby making full use of the characteristics of the amino acid residue itself and the adjacent amino acid residues in the local environment structure diagram, greatly improving the accuracy of interface fusion residue prediction.

[0062] Based on any of the above embodiments, the fusion unit is specifically used for: Determining attention weights of the central node and the connecting nodes based on the initial coding features of the central node and the initial coding features of the connecting nodes; Based on the attention weight, the initial coding feature of the central node and the initial coding features of each connection node are weightedly fused to obtain the fused residue feature.

[0063] Based on any of the above embodiments, the fusion unit is further specifically used for: Based on the graph neural network and the attention weight, the initial coding features of the central node and the initial coding features of each connection node are weightedly fused to obtain an initial fused feature; The initial fusion features are average-pooled to obtain the fusion residue features.

[0064] Based on any of the above embodiments, the protein chain data to be predicted includes: the protein chain sequence to be predicted and the protein chain structure to be predicted; The acquisition unit is specifically used for: Based on the coding model, the sequence characteristics and structural characteristics of each amino acid residue in the protein chain data to be predicted are extracted; The sequence features and structural features of the amino acid residues are spliced ​​to obtain the initial coding features.

[0065] Based on any of the above embodiments, the prediction unit is specifically used for: Inputting the fusion residue features into an interface residue prediction model, and outputting interface residue prediction results of each amino acid residue based on the interface residue prediction model; The interface residue prediction result is a protein interface residue prediction result, a DNA interface residue prediction result, or an RNA interface residue prediction result; The interface residue prediction model is trained based on sample protein complex data.

[0066] Based on any of the above embodiments, the sample protein complex data includes at least one of first protein complex data, second protein complex data and third protein complex data; the first protein complex data is constructed based on proteins and proteins; the second protein complex data is constructed based on proteins and DNA; the third protein complex data is constructed based on proteins and RNA; The device also includes a training unit, which is specifically used for: Obtain initial sample protein complex data; The sample protein complex data are obtained by screening the initial sample protein complex data based on the sequence consistency between the initial sample protein complex data.

[0067] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5 As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the protein interface residue prediction method, which includes: obtaining the protein chain data to be predicted, extracting the initial coding features of each amino acid residue in the protein chain data to be predicted; taking each amino acid residue as a central node and the adjacent residues of the central node as a connecting node, respectively, to construct a local environment structure diagram of each amino acid residue; based on the initial coding features of the central node and the initial coding features of each connecting node, the fusion residue features of the central node are fused; based on the fusion residue features, the interface residue prediction results of each amino acid residue in the protein chain data to be predicted are determined.

[0068] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0069] On the other hand, the present invention also provides a computer program product, which includes a computer program, and the computer program can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the interface residue prediction method of the protein provided by the above-mentioned methods, and the method includes: obtaining the protein chain data to be predicted, and extracting the initial coding features of each amino acid residue in the protein chain data to be predicted; constructing a local environment structure diagram of each amino acid residue with each amino acid residue as a central node and the adjacent residues of the central node as a connecting node; fusing the initial coding features of the central node with the initial coding features of each connecting node to obtain the fused residue features of the central node; and determining the interface residue prediction results of each amino acid residue in the protein chain data to be predicted based on the fused residue features.

[0070] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the interface residue prediction method of the protein provided by the above-mentioned methods, the method comprising: obtaining the protein chain data to be predicted, and extracting the initial coding features of each amino acid residue in the protein chain data to be predicted; constructing a local environment structure diagram of each amino acid residue with each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes; fusing the initial coding features of the central node with the initial coding features of each connecting node to obtain the fused residue features of the central node; and determining the interface residue prediction results of each amino acid residue in the protein chain data to be predicted based on the fused residue features.

[0071] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0072] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0073] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for predicting interface residues of a protein, characterized in that: include: Acquiring the protein chain data to be predicted, and extracting the initial coding features of each amino acid residue in the protein chain data to be predicted; Taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes, a local environment structure diagram of each amino acid residue is constructed; Based on the initial coding feature of the central node and the initial coding features of each connecting node, a fusion residue feature of the central node is obtained; Based on the fusion residue features, the interface residue prediction results of each amino acid residue in the protein chain data to be predicted are determined.

2. The method for predicting interface residues of a protein according to claim 1, characterized in that: The initial coding feature based on the central node is fused with the initial coding features of each connecting node to obtain the fused residue feature of the central node, including: Determining attention weights of the central node and the connecting nodes based on the initial coding features of the central node and the initial coding features of the connecting nodes; Based on the attention weight, the initial coding feature of the central node and the initial coding features of each connection node are weightedly fused to obtain the fused residue feature.

3. The method for predicting interface residues of a protein according to claim 2, characterized in that: The step of weighting and fusing the initial coding features of the central node and the initial coding features of each connection node based on the attention weight to obtain the fused residue features includes: Based on the graph neural network and the attention weight, the initial coding features of the central node and the initial coding features of each connection node are weightedly fused to obtain an initial fused feature; The initial fusion features are average-pooled to obtain the fusion residue features.

4. The method for predicting interface residues of a protein according to any one of claims 1 to 3, characterized in that: The protein chain data to be predicted includes: the protein chain sequence to be predicted and the protein chain structure to be predicted; The extracting and obtaining the initial coding features of each amino acid residue in the protein chain data to be predicted includes: Based on the coding model, the sequence characteristics and structural characteristics of each amino acid residue in the protein chain data to be predicted are extracted; The sequence features and structural features of the amino acid residues are spliced ​​to obtain the initial coding features.

5. The method for predicting interface residues of a protein according to any one of claims 1 to 3, characterized in that: The step of determining the interface residue prediction result of each amino acid residue in the protein chain data to be predicted based on the fusion residue feature comprises: Inputting the fusion residue features into an interface residue prediction model, and outputting interface residue prediction results of each amino acid residue based on the interface residue prediction model; The interface residue prediction result is a protein interface residue prediction result, a DNA interface residue prediction result, or an RNA interface residue prediction result; The interface residue prediction model is trained based on sample protein complex data.

6. The method for predicting interface residues of a protein according to claim 5, characterized in that: The sample protein complex data includes at least one of first protein complex data, second protein complex data, and third protein complex data; The first protein complex data is obtained based on protein-protein construction; The second protein complex data is obtained based on protein and DNA construction; The third protein complex data is obtained based on protein and RNA construction; The step of acquiring the sample protein complex data comprises: Obtain initial sample protein complex data; The sample protein complex data are obtained by screening the initial sample protein complex data based on the sequence consistency between the initial sample protein complex data.

7. A protein interface residue prediction device, characterized in that: include: An acquisition unit, which acquires the protein chain data to be predicted and extracts the initial coding features of each amino acid residue in the protein chain data to be predicted; A graph construction unit, taking each amino acid residue as a central node and the adjacent residues of the central node as connecting nodes, to construct a local environment structure graph of each amino acid residue; A fusion unit, which fuses the initial coding features of the central node with the initial coding features of each connecting node to obtain a fusion residue feature of the central node; The prediction unit determines the interface residue prediction result of each amino acid residue in the protein chain data to be predicted based on the fusion residue feature.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the method for predicting interface residues of a protein according to any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for predicting interface residues of a protein according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for predicting interface residues of a protein according to any one of claims 1 to 6 is implemented.

Images