Protein Interaction Site Prediction Method Based on Local-Global Feature Fusion

Abstract

The present invention provides a method for predicting protein interaction sites based on local-global feature fusion, which relates to the field of protein site prediction and specifically includes the following steps: extracting protein features, including: protein node features and protein edge features; constructing a protein graph structure according to the extracted protein features; inputting the protein graph structure into a graph attention network GAT for further feature extraction to obtain multi-scale features and fusion features; capturing the similarity of the multi-scale features and fusion features through a contrastive loss, inputting the multi-scale features and fusion features into an attention feature fusion module based on Transformer, and finally inputting them into a multi-layer perceptron MLP for classification prediction. The technical solution of the present invention overcomes the problem in the prior art that global information and local information cannot be effectively combined for protein interaction site prediction.

Description

Technical Field

[0001] The present invention relates to the field of protein site prediction, and particularly to a method for predicting protein interaction sites based on local-global feature fusion. Background Art

[0002] Proteins are important components within biological cells and play a crucial role in coordinating various physiological activities. However, proteins often need to interact with other biomolecules to achieve their biological functions, among which protein-protein interaction (PPI) is crucial in numerous biological processes. In these interactions, specific residue regions constitute protein interaction sites (PPIS), which are the interface regions on the protein surface directly involved in binding. Accurately identifying PPIS is of great significance for deeply understanding biological processes, exploring disease mechanisms, and guiding new drug development.

[0003] Information based on protein structure is one of the important strategies for current PPIS prediction and can usually provide higher accuracy. In this method, a protein can be modeled as a graph structure, where amino acids are regarded as nodes in the graph, and edges are determined to exist based on the spatial distance between amino acids. The feature information of proteins is crucial for model performance, but it is often difficult to obtain the optimal prediction result by using only one type of feature. When fusing multiple features, although the prediction accuracy is improved, higher computational complexity will also be introduced. In addition, features such as position-specific scoring matrix (PSSM) rely on large reference databases, resulting in a time-consuming feature extraction process. In recent years, with the development of natural language processing (NLP) technology, protein representation learning methods based on deep learning have gradually emerged, providing new ideas for protein feature extraction.

[0004] Many current methods only regard a protein as a single global graph during modeling, ignoring the hierarchical relationship between the global structure and the local structure. In a real biological system, the local structure of a protein (such as pockets, active sites) often directly determines the binding characteristics, while the global topological structure affects the overall spatial conformation. Therefore, constructing a protein graph only from a global perspective may lose important local information, and only focusing on local features may make it difficult to capture the overall structural relationship. Therefore, how to effectively combine global information and local information in the PPIS prediction task to construct a more refined feature representation remains a key challenge in current research.

[0005] Therefore, there is a need for a method for predicting protein interaction sites that can effectively combine global information and local information. Summary of the Invention

[0006] The main object of the present invention is to provide a method for predicting protein interaction sites based on local-global feature fusion, so as to solve the problem in the prior art that global information and local information cannot be effectively combined for predicting protein interaction sites.

[0007] To achieve the above object, the present invention provides a method for predicting protein interaction sites based on local-global feature fusion, specifically including the following steps:

[0008] S1, extracting protein features, including: protein node features and protein edge features.

[0009] S2, constructing a protein graph structure according to the extracted protein features.

[0010] S3, inputting the protein graph structure into a graph attention network GAT for further feature extraction to obtain multi-scale features and fusion features.

[0011] S4, capturing the similarity of multi-scale features and fusion features through a contrastive loss, inputting the multi-scale features and fusion features into an attention feature fusion module based on Transformer, and finally inputting them into a multi-layer perceptron MLP for classification prediction.

[0012] Furthermore, the protein node features include: evolutionary features, secondary structure features, and physicochemical features of amino acids.

[0013] Furthermore, the evolutionary features include: a position-specific scoring matrix PSSM and a hidden Markov model matrix HMM; normalizing the values in the position-specific scoring matrix PSSM and the hidden Markov model matrix HMM:

[0014] ;

[0015] where represents the original value, and are respectively the minimum value and the maximum value of a certain feature type in the training set, is the value after normalization.

[0016] Furthermore, the secondary structure features are calculated and generated by the DSSP algorithm, and the size of the secondary structure feature matrix is , represents the length of the amino acid sequence, and 14 is the dimension; among them, 9-dimensional features are nine secondary structure states, represented by one-hot encoding; 4-dimensional features are obtained by performing sine and cosine transformations on the torsion angles PHI and PSI of the peptide chain backbone; the last 1-dimensional feature is converted from the solvent accessible surface area SASA to the relative solvent accessibility RSA.

[0017] Furthermore, the physicochemical characteristics of amino acids include: isoelectric point, polarity, pH value, number of hydrogen bond acceptors, number of hydrogen bond donors, octanol-water partition coefficient logP, and topological polar surface area TPSA.

[0018] Furthermore, the extraction of protein edge features in step S1 is specifically as follows:

[0019] The edge features of a protein are represented by the feature matrix of, where 3 represents the dimension; among them, the first dimension is 0 or 1. If there is a direct edge connection between two nodes, it is 1, otherwise it is 0; the second dimension is the position of node and the and Euclidean distance of, as shown in Equation (1); the third dimension is the cosine value of the included angle between and , as shown in Equation (2): the cosine value of , as shown in Equation (2):

[0020] (1);

[0021] (2);

[0022] Among them, is the distance between and , and

[0023] Furthermore, step S2 specifically includes the following steps:

[0024] S2.1, the protein graph structure includes: local graph and global graph. The construction of the adjacency matrices of the local graph and the global graph are shown in Equations (3) and (4) respectively:

[0025] (3);

[0026] (4);

[0027] Among them, , represents the spatial distance between residues, and are the backbone carbon atom coordinates of nodes and respectively, is the adjacency matrix of the local graph, is the adjacency matrix of the global graph, and represent thresholds.

[0028] S2.2, fuse the local graph and global graph information:

[0029] (5);

[0030] Among them, is the adjacency matrix of the fused graph.

[0031] Furthermore, step S3 specifically includes the following steps:

[0032] S3.1, input the protein features extracted in step S1 and the adjacency matrix of the local graph into the first graph attention network GAT to extract the local features of the protein; then use the extracted local features as the initial features of the global graph and input them into the second graph attention network to obtain multi-scale features after convolution.

[0033] S3.2, input the initial protein features and the adjacency matrix of the fused graph into the third graph attention network to extract the fused features of the protein after convolution. The implementations of the first, second, and third graph attention networks are shown in equations (6), (7), and (8):

[0034] (6);

[0035] (7);

[0036] (8);

[0037] Among them, and respectively represent the input feature vectors of node and its neighbor node . represents the edge feature vector between nodes and . represents the concatenation operation. , , and represent the learnable parameter matrices of linear layers at different positions in the graph attention network. represents the activation function. represents the weight of node , and its value range is between 0 and 1. is the attention score between nodes and . represents the updated embedding of node . represents the ReLU activation function.

[0038] Furthermore, step S4 specifically includes the following steps:

[0039] S4.1, concatenate the multi-scale features and the fusion features , that is , and then input the concatenated features into the Transformer attention feature fusion module for processing. The calculation process based on the Transformer encoder is as follows:

[0040] (9);

[0041] Wherein, are the query, key, and value matrices respectively; is the attention mechanism, is the learnable weight matrix, is the dimension of the attention head, is the activation function.

[0042] S4.2, based on the output of the Transformer attention feature fusion module is:

[0043] (10);

[0044] Wherein, is the th attention head, is the output projection matrix, is the concatenation operation.

[0045] S4.3, the final prediction result is:

[0046] (11);

[0047] Wherein, is the normalization layer, is the multi-layer perceptron, is the final prediction result.

[0048] The present invention has the following beneficial effects:

[0049] The present invention innovatively introduces the concepts of global graph and local graph, and respectively extracts the global topological features and local detailed features of proteins through these two graph structures. The contrast loss function is used to capture the similarity between different levels of features, and the two are deeply fused, significantly improving the comprehensiveness and robustness of feature extraction.

[0050] The present invention comprehensively utilizes the sequence features, structural features, and edge features between residues of proteins to construct a multi-dimensional feature representation. This multi-feature fusion strategy not only enriches the feature information but also significantly improves the accuracy of predicting protein interaction sites.

[0051] The present invention proposes a feature fusion mechanism based on Transformer, which can adaptively capture the long-range dependencies between protein sequences and structures and achieve efficient feature fusion through the self-attention mechanism. This mechanism further enhances the model's ability to model complex protein features and provides a more powerful tool for protein function prediction.

[0052] The present invention deeply explores the correlation relationships among protein sequences, structures, and topological information. Meanwhile, it studies the multi-modal feature fusion strategy to enhance the robustness of the model in the case of missing structural information, so as to more effectively utilize the existing data and improve the prediction accuracy and generalization ability. Brief Description of the Drawings

[0053] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required for the description of the specific embodiments or the prior art. Obviously, the drawings in the following description are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings. In the drawings:

[0054] Figure 1 Shows a flowchart of a method for predicting protein interaction sites based on local-global feature fusion of the present invention.

[0055] Figure 2 Shows a schematic diagram of the construction of the local graph and the global graph of the present invention. Detailed Embodiments

[0056] The following will clearly and completely describe the technical solutions of the present invention in conjunction with the drawings. Obviously, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art without creative efforts based on the embodiments of the present invention belong to the scope of protection of the present invention.

[0057] As Figure 1 shown, a method for predicting protein interaction sites based on local-global feature fusion specifically includes the following steps:

[0058] S1, extract protein features, including: protein node features and protein edge features.

[0059] Proteins are composed of amino acid sequences. There are 20 common amino acids, namely alanine (Ala, A), arginine (Arg, R), aspartic acid (Asp, D), asparagine (Asn, N), etc. In the present invention, a protein is regarded as a graph structure, where each amino acid residue is modeled as a node in the graph, and the edges between the nodes are constructed based on the spatial position relationship or covalent bond relationship of the amino acids. Common graph construction methods include based on the adjacency relationship of the amino acid backbone, based on the Euclidean distance threshold between residues, or based on the contact map to define the edge connection method.

[0060] Therefore, before predicting protein interaction sites, it is first necessary to prepare complete protein sequence information and three-dimensional structure information, and construct a reasonable graph structure based on this. On this basis, the present invention further extracts node features and edge features to provide rich information for the training of deep learning models.

[0061] S2. According to the extracted protein features, construct a protein graph structure. The present invention uses AlphaFold3 to generate the structure file of the protein, and constructs the graph structure of the protein based on the Euclidean distance between carbon atoms in the amino acid backbone of the protein. Specifically, the construction of the protein graph depends on the spatial distance between residues , as follows:

[0062]

[0063] Among them, and are the backbone and carbon atom coordinates of amino acid nodes respectively. In order to fully extract the amino acid features of the protein, the present invention constructs three different graph structures, including a global graph, a local graph, and a fusion graph, to depict short-range and long-range topological information, and further enhances the feature expression ability of the protein by combining the contrast learning method.

[0064] S3. Input the protein graph structure into the graph attention network GAT for specific feature extraction to obtain multi-scale features and fusion features.

[0065] S4. Capture the similarity of multi-scale features and fusion features through the contrast loss, input the multi-scale features and fusion features into the attention feature fusion module based on Transformer, and finally input them into the multi-layer perceptron MLP for classification prediction.

[0066] Specifically, the protein node features include: evolutionary features, secondary structure features, and physicochemical features of amino acids.

[0067] Specifically, the evolutionary features include: Position-Specific Scoring Matrix (PSSM) and Hidden Markov Model Matrix (HMM); to ensure the numerical stability of the features, the values in the PSSM and HMM are normalized:

[0068] ;

[0069] where represents the original value, and are the minimum and maximum values of a certain feature type in the training set respectively, is the value after normalization.

[0070] The PSSM is generated by the alignment tool PSI-BLAST v2.10.1 with the number of iterations set to 3 and the E-value set to 0.001, while the HMM matrix is generated by the HHblits v3.0.3 algorithm with default parameters. The shapes of both the PSSM and HMM matrices are where represents the length of the amino acid sequence.

[0071] Specifically, the secondary structure features are calculated by the DSSP algorithm, and the size of the secondary structure feature matrix is where represents the number of amino acids, and 14 is the dimension; among them, 9-dimensional features are nine secondary structure states represented by one-hot encoding; 4-dimensional features are obtained by sine and cosine transformations of the torsion angles PHI and PSI of the peptide chain backbone; the last 1-dimensional feature is converted from the solvent-accessible surface area (SASA) to the relative solvent-accessibility (RSA).

[0072] Specifically, the physicochemical features of amino acids include: isoelectric point, polarity, acidity and alkalinity, number of hydrogen bond acceptors, number of hydrogen bond donors, octanol-water partition coefficient logP, and topological polar surface area (TPSA).

[0073] The secondary structure features are calculated by the DSSP algorithm; the physicochemical features of amino acids include: isoelectric point, polarity, acidity and alkalinity, number of hydrogen bond acceptors, number of hydrogen bond donors, octanol-water partition coefficient logP, and topological polar surface area (TPSA). They form a feature matrix with a shape of . These features comprehensively consider the evolutionary information, structural characteristics, and chemical properties of proteins, providing richer information support for the learning of subsequent models.

[0074] Specifically, the extraction of protein edge features in step S1 is specifically as follows:

[0075] The edge features of the protein are represented by the feature matrix of, where 3 represents the dimension; among them, the first dimension is 0 or 1. If there is a direct edge connection between two nodes, it is 1, otherwise it is 0; the second dimension is the position of node and and and the Euclidean distance of, as shown in formula (1); the third dimension is and and the cosine value of the included angle between, as shown in formula (2):

[0076] (1);

[0077] (2);

[0078] where is and the distance between, is the coordinate initial position.

[0079] Specifically, step S2 specifically includes the following steps:

[0080] S2.1, the protein graph structure includes: a local graph and a global graph. The construction of the adjacency matrices of the local graph and the global graph are shown in formulas (3) and (4) respectively:

[0081] (3);

[0082] (4);

[0083] where , represents the spatial distance between residues, and are respectively the backbone and carbon atom coordinates of nodes , is the adjacency matrix of the local graph, is the adjacency matrix of the global graph, and represent thresholds.

[0084] The present invention sets two thresholds andto construct the local graph and the global graph of the protein, so as to accurately depict the topological relationship between amino acids, such as​Figure 2 As shown. Specifically, when is satisfied , it is considered that there is a local topological relationship between the amino acid nodes and , and a local adjacency matrix is constructed accordingly. This matrix can capture the short-range interaction characteristics of proteins; when , it is considered that there is a global topological relationship between the amino acid nodes and , and a global adjacency matrix is constructed based on this. This matrix can effectively model the correlation information between distant amino acids, thus complementing the deficiency of local information. Through this double-layer topological modeling method, the local and global structural characteristics of proteins can be accurately extracted, enabling the neural network to not only focus on the microscopic amino acid interactions but also capture the macroscopic global topological information, thereby enhancing the expression ability of the model. In addition, this method can enhance the integrity of protein structure information, improve the accuracy of protein function prediction, and provide stronger support for tasks such as protein interaction prediction and function classification.

[0085] S2.2, fusing the local graph and global graph information:

[0086] (5);

[0087] Among them, is the adjacency matrix of the fused graph.

[0088] The fused graph combines the information of the local graph and the global graph, provides a more comprehensive expression of topological relationships, and lays a foundation for calculating the contrast loss in subsequent tasks. When is satisfied , the present invention constructs a fused graph adjacency matrix for the protein, as shown in Equation (5). The fused graph not only integrates the short-range and long-range interaction information but also alleviates to a certain extent the problem of incomplete information that may exist in the local graph and the global graph, enabling the model to further enhance the topological representation ability of proteins based on sequence and structural features, improve the effect of contrast learning, and thus optimize the performance of protein interaction prediction or other downstream tasks.

[0089] Specifically, after the features and the protein graph are constructed, they are input into the graph attention network (GAT) for further feature extraction. Step S3 specifically includes the following steps:

[0090] S3.1, Input the protein features extracted in step S1 and the adjacency matrix of the local graph into the first graph attention network (GAT) to extract the local features of the protein; then use the extracted local features as the initial features of the global graph and input them into the second graph attention network, and obtain multi-scale features after convolution.

[0091] S3.2, Input the initial protein features and the adjacency matrix of the fusion graph into the third graph attention network, and extract the fusion features of the protein after convolution. The implementations of the first, second, and third graph attention networks are shown in equations (6), (7), and (8):

[0092] (6);

[0093] (7);

[0094] (8);

[0095] Among them, and respectively represent the input feature vectors of node and its neighbor node . represents the edge feature vector between node and . represents the concatenation operation. , , and represent the learnable parameter matrices of the linear layers at different positions in the graph attention network. represents the activation function. represents the weight of node , with a value range between 0 and 1, which is obtained by converting its attention score using the function. is the attention score between node and . represents the updated embedding of node . represents the ReLU activation function.

[0096] Specifically, after obtaining the multi-scale features and fusion features in the upstream task, the similarity between the two features is captured through the contrastive loss to make them as close as possible, thereby optimizing the parameters of the upstream model. Subsequently, the two features are input into the attention feature fusion module based on Transformer, and the protein representation ability is further enhanced through feature fusion, and finally input into a multi-layer perceptron (MLP) for classification prediction.

[0097] Step S4 specifically includes the following steps:

[0098] S4.1, concatenate the multi-scale features and the fusion features , that is , and then input the concatenated features into the Transformer attention feature fusion module for processing. The calculation process based on the Transformer encoder is as follows:

[0099] (9);

[0100] Among them, are the query, key, and value matrices respectively; is the attention mechanism, is the learnable weight matrix, is the dimension of the attention head, is the activation function.

[0101] S4.2, based on the output of the Transformer attention feature fusion module is:

[0102] (10);

[0103] Among them, is the th attention head, is the output projection matrix, is the concatenation operation.

[0104] S4.3, the final prediction result is:

[0105] (11);

[0106] Among them, is the normalization layer, is the multi-layer perceptron, is the final prediction result.

[0107] Finally, perform parameter fine-tuning. In the parameter fine-tuning and prediction stage, the present invention uses 80% of the protein data for training, 20% of the data for testing, and adopts five-fold cross-validation to evaluate the stability and generalization ability of the model. By adjusting the learning rate, batch size, optimizer (such as AdamW), and regularization parameters (such as Dropout and weight decay), the model is fine-tuned to optimize its performance. During the training process, the early stopping strategy is used to prevent overfitting, and the accuracy, F1 score, etc. of the model are evaluated on the test set. Finally, the reliability of the model is verified through the results of five-fold cross-validation.

[0108] Of course, the above description is not a limitation of the present invention, and the present invention is not limited to the above examples. Changes, modifications, additions, or substitutions made by those skilled in the art within the scope of the essence of the present invention should also fall within the protection scope of the present invention.

Claims

1. A protein interaction site prediction method based on local-global feature fusion, characterized in that: The specific steps include: S1, extract protein features, including protein node features and protein edge features; S2, constructing the protein graph structure based on the extracted protein features; S3, input the protein graph structure into the graph attention network GAT for further feature extraction to obtain multi-scale features and fusion features; S4, captures the similarity of multi-scale features and fused features through contrast loss, inputs the multi-scale features and fused features into the attention feature fusion module based on Transformer, and finally inputs them into the multi-layer perceptron MLP for classification prediction; Step S2 specifically includes the following steps: S2.1, the protein graph structure includes: local graph and global graph. The adjacency matrix of local graph and global graph is constructed as shown in equations (3) and (4) respectively: （3）； （4）； in, , represents the spatial distance between residues, and Node and Main chain Carbon atom coordinates, is the adjacency matrix of the local graph, is the adjacency matrix of the global graph, and represents the threshold value; S2.2, fusion of local graph and global graph information: （5）； in, is the adjacency matrix of the fused graph.

2. A protein interaction site prediction method based on local-global feature fusion according to claim 1, characterized in that: Protein node characteristics include: evolutionary characteristics, secondary structure characteristics and physicochemical characteristics of amino acids.

3. A protein interaction site prediction method based on local-global feature fusion according to claim 2, characterized in that: The evolutionary features include: a position-specific score matrix PSSM and a hidden Markov model matrix HMM; the values ​​in the position-specific score matrix PSSM and the hidden Markov model matrix HMM are normalized: ； in, Represents the original value, and are the minimum and maximum values ​​of a feature type in the training set, respectively. is the normalized value.

4. A protein interaction site prediction method based on local-global feature fusion according to claim 2, characterized in that: The secondary structure features are calculated by the DSSP algorithm, and the size of the secondary structure feature matrix is , Represents the length of the amino acid sequence, and 14 is the dimension; among them, the 9-dimensional features are nine secondary structure states, represented by one-hot encoding; the 4-dimensional features are obtained by sine and cosine transforming the torsion angles PHI and PSI of the peptide chain main chain; the last 1-dimensional feature is converted from the solvent accessible surface area SASA to the relative solvent accessibility RSA.

5. The method for predicting protein interaction sites based on local-global feature fusion according to claim 2, characterized in that: The physicochemical characteristics of amino acids include isoelectric point, polarity, pH, number of hydrogen bond acceptors, number of hydrogen bond donors, octanol-water partition coefficient logP and topological polar surface area TPSA.

6. A protein interaction site prediction method based on local-global feature fusion according to claim 1, characterized in that: The protein edge features extracted in step S1 are specifically: The protein edge features are The feature matrix is ​​represented by , where 3 represents the dimension; the first dimension is 0 or 1, if two nodes are directly connected by an edge, it is 1, otherwise it is 0; the second dimension is the node and Location and The Euclidean distance , as shown in formula (1); the third dimension is and Angle between The cosine value of , as shown in formula (2): （1）； （2）； in, for and The distance between is the initial coordinate position.

7. The method for predicting protein interaction sites based on local-global feature fusion according to claim 1, characterized in that: Step S3 specifically includes the following steps: S3.1, input the protein features extracted in step S1 and the adjacency matrix of the local graph into the first graph attention network GAT1 to extract the local features of the protein; then use the extracted local features as the initial features of the global graph and input them into the second graph attention network GAT2 to obtain multi-scale features after convolution; S3.2, the initial protein features and the adjacency matrix of the fusion graph are input into the third graph attention network GAT3, and the fusion features of the protein are extracted after convolution. The implementation of the first, second, and third graph attention networks is shown in equations (6), (7), and (8): （6）； （7）； （8）； in, and Respectively represent nodes and its neighbor nodes The input feature vector is Representation Node and The edge feature vector between Represents a splicing operation, , , and represents the learnable parameter matrix of the linear layer at different positions in the graph attention network, represents the activation function, Representation Node The weight of is between 0 and 1. For Node and The attention score between Representation Node Updated embed, Represents the ReLU activation function.

8. The method for predicting protein interaction sites based on local-global feature fusion according to claim 1, characterized in that: Step S4 specifically includes the following steps: S4.1, multi-scale features and fusion features To splice, , and then the concatenated features The input is processed based on the Transformer attention feature fusion module. The calculation process based on the Transformer encoder is as follows: （9）； in, are query, key, and value matrices respectively; is the attention mechanism, is the learnable weight matrix, is the dimension of the attention head, is the activation function; S4.2, Output of Transformer-based Attention Feature Fusion Module for: （10）； in, It is Attention head, is the output projection matrix, For connection operation; S4.3, the final prediction result is: （11）； in, is the normalization layer, is a multi-layer perceptron, is the final prediction result.

Images