Protein interaction prediction method based on cross graph representation learning

Abstract

The invention discloses a protein interaction prediction method based on cross graph representation learning, and belongs to the field of protein interaction, and the method comprises the following steps: S1, collecting a prediction data set of protein interaction, carrying out the feature processing of proteins, and constructing a protein graph; s2, learning space structure information of the protein map by using a GCN-based map encoder; s3, using an encoder based on a self-attention module to learn information between receptor protein and ligand protein; s4, learning information of residues between the receptor protein and the ligand protein by using a dual interaction graph module; and S5, generating a classifier for protein interaction prediction. The method is used for solving the problem that the performance of an existing protein interaction prediction method is often degraded in different fields.

Description

Technical Field

[0001] The present invention relates to the field of protein-protein interaction, and in particular to a protein-protein interaction prediction method based on cross-graph representation learning. Background Art

[0002] Protein interaction prediction (PPI) is to determine whether two proteins can interact with each other through various technical means. Proteins are the basic substances of all life, and the interaction between proteins is essential to life. Protein interaction prediction is involved in almost all biological processes, including metabolism, genetic pathways, and signal cascades, among which they are used for DNA replication and transcription, RNA translation, energy production, signal transduction, immunity, etc. The interaction between proteins determines the mechanism of molecules and cells and participates in most biological processes in organisms. It has important applications in explaining pathogenic variants, developing drugs targeting PPIs, and designing protein binders to regulate protein functions. There are many methods for studying protein interaction prediction:

[0003] 1) Based on experimental analysis methods, various experimental analyses have been widely used for protein interaction prediction (PPI) identification. PPI detection methods include widely used low-throughput methods such as co-immunoprecipitation (Co-IP) and bioluminescence resonance energy transfer (BRET), as well as a series of high-throughput methods such as yeast two-hybrid (Y2H) screening. X-ray crystallography and nuclear magnetic resonance (NMR) are commonly used methods for determining the structure of protein complexes. However, the complexity of PPI makes experimental analysis time-consuming, laborious, and very expensive.

[0004] 2) Based on machine learning methods, machine learning algorithms such as support vector machines and random forests are used to extract features of protein sequences and structures and train them to establish protein interaction prediction models. This type of method can be calculated relatively simply, can automatically learn the relationship between proteins, and has a certain adaptability to different individuals and environments. However, the disadvantage is that the quality and selection of feature extraction have a great impact on the results, and a large amount of labeled data is required for training.

[0005] 3) Methods based on deep learning. Deep learning has powerful nonlinear transformation capabilities and has received widespread attention in various fields. Algorithms based on deep learning are increasingly being used in the prediction of protein interactions. Compared with machine learning, deep learning methods can automatically extract effective features from large-scale raw data sets without prior knowledge. Although deep learning methods have made great progress in PPI prediction, there are still some problems. First of all, many methods are mostly based on recursive neural networks or convolutional neural networks to capture the characteristics of primary sequences in one-dimensional space, while ignoring the spatial information of protein molecules. Secondly, the area where proteins interact with each other accounts for a relatively small part of the protein, so fine-grained modeling is quite necessary for PPI prediction, which has not been fully studied in previous studies. Summary of the invention

[0006] The purpose of the present invention is to provide a protein interaction prediction method based on cross-graph representation learning to solve the performance degradation problem often exhibited by existing protein interaction prediction methods in different fields.

[0007] To achieve the above object, the present invention provides a protein interaction prediction method based on cross-graph representation learning, comprising the following steps:

[0008] S1, collects prediction data sets of protein interactions, then performs feature processing on proteins and constructs protein graphs;

[0009] S2, using GCN-based graph encoder to learn the spatial structure information of protein graphs;

[0010] S3, uses an encoder based on a self-attention module to learn the information between receptor proteins and ligand proteins;

[0011] S4, using the dual interaction graph module to learn the information of residues between receptor protein and ligand protein;

[0012] S5, Generating classifiers for protein interaction predictions.

[0013] Preferably, in S1, the prediction data set needs to be cleaned to remove too short sequences and too long sequences in the protein sequence.

[0014] Preferably, in S1, the specific steps of constructing the protein map are:

[0015] S11, input the cleaned protein sequence into the ESM2 pre-training model;

[0016] S12, a series of protein sequences are feature embedded through the ESM2 pre-training model to obtain a series of residue embedding representations;

[0017] S13, which measures the similarity between residues and determines the relationship between the edges of residues, is defined as follows:

[0018]

[0019] Among them, Z l is the characteristic embedding of ligand protein, M l (Z l ) is the relationship matrix between residue sequences;

[0020] S14, for the input protein pair, we use residues as nodes and their edge relationships as defined in S13 to construct a protein graph:

[0021]

[0022] Among them, G l Represents the ligand protein map, Z l represents the feature embedding of ligand protein, M l Represents the edge relationship between nodes;

[0023] Finally, repeat steps S11 to S14 to obtain the receptor protein map: (G r (Z r ,M r )).

[0024] Preferably, the specific process of S2 is:

[0025] Use GCN to learn the representation of nodes based on their neighbor nodes, and assign weights according to the distance between the neighbor nodes and the central node. The graph encoding process is as follows:

[0026] First, the adjacency matrix is ​​normalized. The normalization process is to add self-loops to the adjacency matrix (each node is connected to itself) and use the degree matrix for normalization:

[0027]

[0028] Where I is the identity matrix, which represents the self-loops connecting each node to itself. is the semi-inverse of the degree matrix;

[0029] Secondly, feature aggregation is performed. For each layer of GCN, the normalized adjacency matrix Aggregate the features of the nodes and represent the features of each node in the graph at the kth layer as X (k) , the node feature representation of the k+1th layer is X (k+1) Then calculate it using the following formula:

[0030]

[0031] Where X(k) represents the node feature matrix of the kth layer, W (k) represents the weight matrix of the kth layer, represents the normalized adjacency matrix, σ represents the activation function, and the activation function used is RELU;

[0032] Then, the graph encoder is used to learn the ligand protein graph and the receptor protein graph to obtain the feature vectors of the ligand protein. and the feature vector within the receptor protein The feature vector of ligand protein For example, the formula is as follows:

[0033]

[0034] Where softmax(Z l ) represents the features of ligand proteins generated by the graph encoder.

[0035] Preferably, the specific process of S3 is to use the feature vector in the ligand protein As an example, the ligand protein sequence is input into the self-attention model to obtain the contextual representation of each residue:

[0036]

[0037]

[0038] where Q, K and V represent the query, key and value vectors of the ligand protein, They represent the corresponding weights of the neural network initialization, d K represents the dimension of the feature vector K;

[0039] The residue context representation of the receptor protein was obtained in the same manner.

[0040] Preferably, the dual interaction graph module includes a normalization layer, a multi-head cross attention module, a residual connection and a feedforward neural network layer, and the multi-head cross attention module is as follows:

[0041]

[0042]

[0043] Q r is the query vector of the receptor protein, K l , V l is the key and value vector of ligand protein, and m is the number of cross-attention modules;

[0044] Then, a feedforward neural network is used to learn the sequence features to obtain the receptor protein feature vector and ligand protein feature vector The ligand protein feature vector The way to obtain is as follows:

[0045]

[0046] in Represents the characteristics of ligand proteins generated by protein-protein interaction modules, Represents the feature information learned by the protein-protein representation learning module;

[0047] Similarly, the feature vector in the receptor protein Input the feedforward neural network for learning to obtain the receptor protein feature vector

[0048] Preferably, in S5, a learnable parameter λ is set to fuse receptor protein information and ligand protein information.

[0049]

[0050]

[0051] Among them, H l Indicates the ligand protein information after fusion, H r Indicates the receptor protein information after fusion;

[0052] Then H l and H r Put it together and use the fully connected layer as a classifier for protein interactions with a binary cross entropy loss as the target:

[0053]

[0054]

[0055] in, is the number of training samples;

[0056] The probability of contact between each residue between proteins is predicted by fusing the receptor protein information and the ligand protein information.

[0057] Therefore, the present invention adopts a protein interaction prediction method based on cross-graph representation learning of the above structure, which has the following advantages:

[0058] 1) Considering that proteins have a certain spatial structure, residues are used as nodes, and the similarities between nodes are used to construct edge information and a graph model, which increases the spatial information of proteins.

[0059] 2) In order to learn the contextual information of protein sequences, a self-attention encoder is used to learn the long-range dependencies between proteins.

[0060] 3) A dual interaction graph module is proposed to learn fine-grained representations between proteins.

[0061] The technical solution of the present invention is further described in detail below through the accompanying drawings and embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] Figure 1 A technical network diagram of a protein interaction prediction method based on cross-graph representation learning of the present invention;

[0063] Figure 2 A schematic diagram of a process of a protein interaction prediction method based on cross-graph representation learning of the present invention;

[0064] Figure 3 The effectiveness of the number of graph convolution layers of the protein intra-representation learning module in a protein interaction prediction method based on cross-graph representation learning of the present invention is compared;

[0065] Figure 4 This invention provides a comparison of the protein interaction prediction method based on cross-graph representation learning and the influence of the number of interaction graph module layers on protein representation learning. DETAILED DESCRIPTION

[0066] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings here can be arranged and designed in various different configurations.

[0067] Therefore, the following detailed description of the embodiments of the present invention provided in the accompanying drawings is not intended to limit the scope of the invention claimed for protection, but merely represents selected embodiments of the present invention. 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.

[0068] It should be noted that similar reference numerals and letters denote similar items in the following drawings, and therefore, once an item is defined in one drawing, further definition and explanation thereof is not required in subsequent drawings.

[0069] In the description of the present invention, it should be noted that the terms "upper", "lower", "inside", "outside", etc. indicate directions or positional relationships based on the directions or positional relationships shown in the accompanying drawings, or are directions or positional relationships in which the product of the invention is usually placed when in use. They are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present invention.

[0070] In the description of the present invention, it is also necessary to explain that, unless otherwise clearly specified and limited, the terms "setting", "installation" and "connection" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0071] Some embodiments of the present invention are described in detail below in conjunction with the accompanying drawings. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0072] In the description of the present invention, it should be noted that the directions or positional relationships indicated by terms such as “upper”, “lower”, “left”, “right”, “inside” and “outside” are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present invention and simplifying the description. They do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction. Therefore, they cannot be understood as limitations on the present invention.

[0073] The specific model specifications need to be selected and determined based on the actual specifications of the device, and the specific selection calculation method adopts the existing technology in this field, so it will not be described in detail.

[0074] Example

[0075] like Figure 1 , Figure 2 As shown, the following steps are included:

[0076] S1, collects prediction data sets of protein interactions, then performs feature processing on proteins and constructs protein graphs;

[0077] In S1, the prediction data set needs to be cleaned to remove too short and too long sequences in the protein sequence.

[0078] The specific steps of S1 protein map construction are:

[0079] S11, input the cleaned protein sequence into the ESM2 pre-training model;

[0080] ESM2 is an unsupervised Transformer model used as a pre-training model. It is trained on protein sequences in the UniRef database. ESM2 only has a simple masked language modeling objective during training, but ESM2 has effectively achieved success in various tasks by leveraging the diversity of protein sequences. Using the ESM2 model as a pre-training model, rich protein embedding information (i.e., representation information in high-dimensional space) is provided for each protein sequence. These embeddings implicitly give structural information about each protein, generating 2560-dimensional embeddings for each protein sequence. Used for the construction of ligand and receptor protein graph models.

[0081] S12, a series of protein sequences are feature embedded through the ESM2 pre-training model to obtain a series of residue embedding representations;

[0082] S13, which measures the similarity between residues and determines the relationship between the edges of residues, is defined as follows:

[0083]

[0084] Among them, Z l is the characteristic embedding of ligand protein, M l (Z l ) is the relationship matrix between residue sequences;

[0085] S14, for the input protein pair, we use residues as nodes and their edge relationships as defined in S13 to construct a protein graph:

[0086]

[0087] Among them, G l Represents the ligand protein map, Z l represents the feature embedding of ligand protein, M l Represents the edge relationship between nodes;

[0088] Finally, repeat steps S11 to S14 to generate the receptor protein map: (G r (Z r ,M r )).

[0089] S2, using GCN-based graph encoder to learn the spatial structure information of protein graphs;

[0090] The specific process of S2 is:

[0091] Use GCN to learn the representation of nodes based on their neighbor nodes, and assign weights according to the distance between the neighbor nodes and the central node. The graph encoding process is as follows:

[0092] First, the adjacency matrix is ​​normalized to avoid the influence of degree differences in the graph on the convolution operation. The normalization method is to add self-loops to the adjacency matrix (each node is connected to itself) and use the degree matrix for normalization:

[0093]

[0094] Where i is the identity matrix, representing the self-loops connecting each node to itself. is the semi-inverse of the degree matrix.

[0095] Secondly, feature aggregation is performed. For each layer of GCN, the normalized adjacency matrix Aggregate node features. The features of each node in the graph at the kth layer are represented as X (k) Then the node feature representation X at the k+1th layer (k+1) It can be calculated by the following formula:

[0096]

[0097] Where X (k) represents the node feature matrix of the kth layer, W (k) represents the weight matrix of the kth layer, Represents the normalized adjacency matrix, σ represents the activation function, and the RELU (Rectified Linear Unit) activation function is commonly used here.

[0098] In the GCN-based graph encoding module, neighbor information is aggregated through the adjacency structure of the graph, and the influence of node degree differences is eliminated by normalizing the adjacency matrix. Finally, multiple convolutional layers are stacked to extract more complex graph structure features, and the network parameters are updated through back propagation. Then, the graph encoder is used to learn the ligand protein graph and the receptor protein graph, and the feature vectors of the ligand protein are obtained respectively. and the feature vector within the receptor protein The feature vector of ligand protein For example, the formula is as follows:

[0099]

[0100] Where softmax(Z l ) represents the features of ligand proteins generated by the graph encoder.

[0101] S3, uses an encoder based on a self-attention module to learn the information between receptor proteins and ligand proteins;

[0102] The specific process of S3 is to use the feature vector in the ligand protein As an example, the ligand protein sequence is input into the self-attention model to obtain the contextual representation of each residue:

[0103]

[0104]

[0105] where Q, K and V represent the query, key and value vectors of the ligand protein, They represent the corresponding weights of the neural network initialization, d K represents the dimension of the feature vector K;

[0106] The residue context representation of the receptor protein was obtained in the same manner.

[0107] For the feature vector of the protein pair obtained in the previous module and First, a protein feature encoder is introduced to learn the information between the contexts of a single protein residue. Then a protein interaction module is designed to learn the information between two proteins in a fine-grained manner.

[0108] S4, using the dual interaction graph module to learn the information of residues between receptor protein and ligand protein;

[0109] The dual interaction graph module includes a normalization layer, a multi-head cross-attention module, a residual connection, and a feedforward neural network layer. The multi-head cross-attention module is as follows:

[0110]

[0111]

[0112] Q r is the query vector of the receptor protein, K l , V l is the key and value vector of ligand protein, and m is the number of cross-attention modules;

[0113] Then, a feedforward neural network is used to learn the sequence features to obtain the receptor protein feature vector and ligand protein feature vector The ligand protein feature vector The way to obtain is as follows:

[0114]

[0115] in Represents the characteristics of ligand proteins generated by protein-protein interaction modules, Represents the feature information learned by the protein-protein representation learning module;

[0116] Similarly, the feature vector in the receptor protein Input the feedforward neural network for learning to obtain the receptor protein feature vector

[0117] When the receptor protein is the query vector, the ligand protein is used as the key and value vectors, and the residues of the receptor protein learn fine-grained representations from the ligand protein. The module consists of a normalization layer, a multi-head criss-cross attention module, a residual connection, and a feedforward neural network layer. Residual connections are adopted around the multi-head criss-cross attention module and the feedforward layer. The residual connection can effectively learn the residual, thereby preventing the gradient vanishing problem. A feedforward neural network layer is performed to represent the attention passed through the multi-head criss-cross attention module. In the case of the multi-head criss-cross attention module, it can be extended from the m-th single-head criss-cross attention module. Then, a feedforward neural network is adopted to learn the sequence features. The feedforward neural network consists of two fully connected layers and an activation function to perform nonlinear transformations and map the representation of each position.

[0118] S5, Generating classifiers for protein interaction predictions.

[0119] In S5, the learnable parameter λ is set, and the receptor protein information and the ligand protein information are fused.

[0120]

[0121]

[0122] Among them, H l Indicates the ligand protein information after fusion, H r Indicates the receptor protein information after fusion;

[0123] Then H l and H r Put it together and use the fully connected layer as a classifier for protein interactions with a binary cross entropy loss as the target:

[0124]

[0125]

[0126] in, is the number of training samples;

[0127] The probability of contact between each residue between proteins is predicted by fusing the receptor protein information and the ligand protein information.

[0128] The present invention was verified on yeast and H. sapiens, commonly used datasets for protein interaction prediction, and compared with other benchmark methods. Finally, the effectiveness of the module was analyzed.

[0129] Experimental setup

[0130] Prediction datasets: The prediction datasets include yeast and H. sapiens datasets. The intraspecies PPI dataset of yeast is a widely used benchmark dataset in state-of-the-art methods. It contains 2497 proteins, of which 5594 positive samples and 5594 negative samples. The high-quality yeast dataset is extracted from the interacting protein database and contains only the most reliable physical interactions. Random pairing of proteins without evidence of interaction will produce negative interactions. The H. sapiens dataset contains 15816 proteins, of which 38345 positive samples and 383450 negative samples.

[0131] Model experiment details: Specific details: For the input of sequence information of protein pairs, a pre-trained model (ESM2) is used to generate 2560-dimensional features. For the intra-protein representation learning module. A graph encoder is used, which consists of a three-layer GCN module to generate 512-dimensional feature representations for intra-protein learning. For the inter-protein representation learning module. A self-attention module is used as an encoder to learn the contextual information of the sequence. Then a multi-head cross-attention module is used to learn the information between sequences. Finally, the residue representations of the protein pairs are connected and sent to the fully connected layer for binary prediction. Rectified linear unit (ReLU) is used as the activation function, and dropout is set to 0.4 to avoid overfitting. A model trained with ADAM for 100 cycles is used on a NVIDIA RTX 4090 GPU card with 24GB memory, with a learning rate of 0.0001. The model is trained in an end-to-end manner using cross entropy loss.

[0132] Experimental Results

[0133] The performance comparison of the model with the baseline method on the yeast dataset is shown below.

[0134]

[0135] In the table, the model of the present invention is compared with several state-of-the-art methods on yeast datasets, namely MCD-SVM, RF-LPQ, KNN-CTD, EELM-PCA, DeepPPI, SAE, DPPI, DNN-PPI, PIPR and TAGPPI. This clearly shows that the method of the present invention achieves new state-of-the-art performance and obtains significant improvements on all yeast datasets. Specifically, the performance of the method of the present invention is 0.9798 on ACC, 0.9812 on Pre, 0.9785 on Sen, 0.9812 on Spe, 0.9798 on F1, 0.9597 on MCC, and 0.9798 on AUC. Among all the indicators, six indicators reached the highest and one indicator ranked second. All of the above results demonstrate the effectiveness of the framework proposed in the present invention. The gains are attributed to the proposed information learning within the protein and the information propagation of inter-protein learning during the learning process, which better captures the encoding of the relationship between residues.

[0136] Experimental results analysis

[0137] Our invention achieves superior performance compared to existing state-of-the-art methods, so in addition to the performance evaluation, additional experiments are conducted to clarify how each module contributes to the classification task and the sensitivity of the hyperparameters used in the method.

[0138] like Figure 3 As shown in Figure 2, for a comprehensive comparison, experiments are conducted using multiple graph convolutional layers in the graph encoder learned within the graph. It can be observed that the accuracy of stacking three convolutional layers is higher than that of other cases. In addition, when only one convolutional layer is used, the aggregated information of the neighbors is not enough. In addition, more graph convolutional layers lead to degraded performance due to over-smoothing.

[0139] like Figure 4 As shown in the figure, in order to study the influence of the interaction graph module in our framework, multiple layers are stacked and the performance between them is compared. The performance varies with the number of interaction graph module layers. When the number of layers is set to 0, it means that the information of the ligand and receptor proteins will not propagate to each other, and the accuracy is significantly reduced. On the contrary, when two layers are combined, the best accuracy can be obtained. This shows that the information from another protein effectively improves the residue representation and promotes the recognition of residue pair relationships.

[0140] Therefore, the present invention adopts a protein interaction prediction method based on cross-graph representation learning of the above structure, which solves the performance degradation problem often exhibited by existing protein interaction prediction methods in different fields.

[0141] Finally, it should be noted that the above embodiments are only used to illustrate the technical solution of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that they can still modify or replace the technical solution of the present invention with equivalents, and these modifications or equivalent replacements cannot cause the modified technical solution to deviate from the spirit and scope of the technical solution of the present invention.

Claims

1. A protein interaction prediction method based on cross-graph representation learning, characterized in that: The following steps are involved: S1, collects prediction data sets of protein interactions, then performs feature processing on proteins and constructs protein graphs; S2, using GCN-based graph encoder to learn the spatial structure information of protein graphs; S3, uses an encoder based on a self-attention module to learn the information between receptor proteins and ligand proteins; S4, using the dual interaction graph module to learn the information of residues between receptor protein and ligand protein; S5, Generating classifiers for protein interaction predictions.

2. A protein interaction prediction method based on cross-graph representation learning according to claim 1, characterized in that: In S1, the prediction data set needs to be cleaned to remove too short and too long sequences in the protein sequence.

3. A protein interaction prediction method based on cross-graph representation learning according to claim 2, characterized in that: In S1, the specific steps of constructing the protein graph are: S11, input the cleaned protein sequence into the ESM2 pre-training model; S12, a series of protein sequences are feature embedded through the ESM2 pre-training model to obtain a series of residue embedding representations; S13, which measures the similarity between residues and determines the relationship between the edges of residues, is defined as follows: Among them, Z l is the characteristic embedding of ligand protein, M l (Z l ) is the relationship matrix between residue sequences; S14, for the input protein pair, the residues are used as nodes, and the edge relationships are defined as in S13 to construct a protein graph: Among them, G l Represents the ligand protein map, Z l represents the feature embedding of ligand protein, M l Represents the edge relationship between nodes; Finally, repeat steps S11 to S14 to obtain the receptor protein map: (G r (Z r ,M r )).

4. The protein interaction prediction method based on cross-graph representation learning according to claim 3, characterized in that: The specific process of S2 is: Use GCN to learn the representation of nodes based on their neighbor nodes, and assign weights according to the distance between the neighbor nodes and the central node. The graph encoding process is as follows: First, the adjacency matrix is ​​normalized. The normalization process is to add self-loops to the adjacency matrix and use the degree matrix for normalization: Where I is the identity matrix, which represents the self-loops connecting each node to itself. is the semi-inverse of the degree matrix; Secondly, feature aggregation is performed. For each layer of GCN, the normalized adjacency matrix Aggregate the features of the nodes and represent the features of each node in the graph at the kth layer as X (k) , the node feature representation of the k+1th layer is X (k+1) Then calculate it using the following formula: Where X (k) represents the node feature matrix of the kth layer, W (k) represents the weight matrix of the kth layer, represents the normalized adjacency matrix, σ represents the activation function, and the activation function used is RELU; Then, the graph encoder is used to learn the ligand protein graph and the receptor protein graph to obtain the feature vectors of the ligand protein. and the feature vector within the receptor protein The feature vector of ligand protein For example, the formula is as follows: Where softmax(Z l ) represents the features of ligand proteins generated by the graph encoder.

5. The protein interaction prediction method based on cross-graph representation learning according to claim 4, characterized in that: The specific process of S3 is to use the feature vector in the ligand protein As an example, the ligand protein sequence is input into the self-attention model to obtain the contextual representation of each residue: where Q, K and V represent the query, key and value vectors of the ligand protein, They represent the corresponding weights of the neural network initialization, d K represents the dimension of the feature vector K; The residue context representation of the receptor protein was obtained in the same manner.

6. A protein interaction prediction method based on cross-graph representation learning according to claim 5, characterized in that: The dual interaction graph module includes a normalization layer, a multi-head cross-attention module, a residual connection, and a feedforward neural network layer. The multi-head cross-attention module is as follows: Q r is the query vector of the receptor protein, k l , V l is the key and value vector of ligand protein, and m is the number of cross-attention modules; Then, a feedforward neural network is used to learn the sequence features to obtain the receptor protein feature vector and ligand protein feature vector The ligand protein feature vector The way to obtain is as follows: in Represents the characteristics of ligand proteins generated by protein-protein interaction modules, Represents the feature information learned by the protein-protein representation learning module; Similarly, the feature vector in the receptor protein Input the feedforward neural network for learning to obtain the receptor protein feature vector 7. A protein interaction prediction method based on cross-graph representation learning according to claim 6, characterized in that: In S5, the learnable parameter λ is set, and the receptor protein information and ligand protein information are fused. Among them, H l Indicates the ligand protein information after fusion, H r Represents the receptor protein information after fusion; Then H l and H r Put it together and use the fully connected layer as a classifier for protein interactions with a binary cross entropy loss as the target: in, is the number of training samples; The probability of contact between each residue between proteins is predicted by fusing the receptor protein information and the ligand protein information.

Images