Epidermal growth factor receptor pathogenicity prediction method and device, server and medium

Abstract

The invention discloses an epidermal growth factor receptor pathogenicity prediction method, an epidermal growth factor receptor pathogenicity prediction device, a server and a medium, and the epidermal growth factor receptor pathogenicity prediction method comprises the following steps: extracting three-dimensional space conformation amino acid residue characteristics of a compound mutation amino acid sequence, and generating an initial amino acid residue structure diagram; predicting by using the first graph neural network, and generating an amino acid cluster structure graph of a second level through weighted polymerization; carrying out prediction by utilizing a second graph neural network, and carrying out weighted polymerization to generate a higher-level protein subunit structure graph; and predicting by using a third graph neural network, and then carrying out feature mapping to obtain a protein structure feature vector for pathogenicity evaluation of a mutation site. Compared with the prior art, the method has the advantages that a layer-by-layer deep graph neural network architecture is adopted, and the comprehensive influence of a plurality of mutation sites on a multi-level protein structure is accurately captured, so that the pathogenicity of the mutation sites of the epidermal growth factor receptor is effectively evaluated, and the prediction accuracy and comprehensiveness are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of epidermal growth factor receptors, and in particular to a method, device, server and medium for predicting pathogenicity of epidermal growth factor receptors. Background Art

[0002] Epidermal growth factor receptor (EGFR) is an important member of the tyrosine kinase receptor family and plays an important role in maintaining cell physiological functions. However, EGFR is easily affected by multiple factors such as environment, genetics, and lifestyle habits and mutates, becoming a pathogenic cause of many cancers and posing a threat to human health. Currently, tyrosine kinase inhibitors (TKIs) are the main targeted drugs for cancer patients with EGFR mutations, which control tumor growth by inhibiting EGFR activity. However, the diversity and continuous dynamic changes of EGFR mutations pose challenges to targeted therapy.

[0003] By predicting potential mutation sites of EGFR and evaluating their pathogenicity, more accurate decision support can be provided for clinical treatment. Currently, models such as graph neural networks are used to predict the protein structure of a single mutation site and evaluate its pathogenicity. However, when predicting protein structural variations caused by compound mutations containing multiple mutation sites, it is impossible to accurately capture and predict the spatial position relationship between multiple mutation sites and their comprehensive impact on protein structure, and thus it is impossible to accurately evaluate the pathogenicity of the mutation site. Summary of the invention

[0004] The embodiments of the present invention provide a method, device, server and medium for predicting the pathogenicity of epidermal growth factor receptor, which can accurately predict the impact of multiple EGFR mutation sites on protein structure, thereby accurately evaluating the pathogenicity of the mutation sites.

[0005] In a first aspect, an embodiment of the present invention provides a method for predicting pathogenicity of epidermal growth factor receptor, comprising:

[0006] Extracting amino acid residue features based on the three-dimensional spatial conformation of a composite mutant amino acid sequence containing multiple mutation sites, wherein the amino acid residue features include: amino acid residue type features, amino acid residue centroid coordinate features, key residue features and dihedral angle features, and the amino acid residue features are in matrix form;

[0007] According to the amino acid residue characteristics, an initial amino acid residue structure graph is generated, wherein the characteristics of each node in the initial amino acid residue structure graph are the corresponding amino acid residue characteristics; the initial amino acid residue structure graph is input into the trained first graph neural network, and the amino acid residue characteristics of each node are updated using the convolution layer of the first graph neural network to obtain a predicted amino acid residue feature matrix for each node;

[0008] According to the predicted amino acid residue feature matrix of each node, DBSCAN clustering is performed on the amino acid residues based on spatial proximity, and all predicted amino acid residue feature matrices in each category are weighted aggregated to generate the amino acid cluster feature matrix of each category;

[0009] According to the amino acid cluster feature matrix, an initial amino acid cluster structure diagram is generated, wherein the feature of each node in the initial amino acid cluster structure diagram is the corresponding amino acid cluster feature matrix; the initial amino acid cluster structure diagram is input into the trained second graph neural network, and the amino acid cluster feature matrix of each node is updated by using the convolution layer of the second graph neural network to obtain a predicted amino acid cluster feature matrix of each node;

[0010] According to the predicted amino acid cluster feature matrix, the amino acid cluster combination threshold is determined, the predicted amino acid cluster feature matrix is ​​hierarchically clustered using the amino acid cluster combination threshold, all predicted amino acid cluster feature matrices in each category are weightedly aggregated to generate multiple protein subunit feature matrices;

[0011] According to the protein subunit feature matrix, an initial protein subunit structure graph is generated, wherein the feature of each node in the initial protein subunit structure graph is the corresponding protein subunit feature matrix; the initial protein subunit structure graph is input into the trained third graph neural network, and the protein subunit feature of each node is updated by using the convolution layer of the third graph neural network to obtain a predicted protein subunit feature matrix of each node;

[0012] All predicted protein subunit feature matrices are feature mapped to generate predicted protein structure feature vectors, which are then used to evaluate the pathogenicity of mutation sites.

[0013] Furthermore, the amino acid residues are clustered based on spatial proximity according to the predicted amino acid residue feature matrix of each node, and weighted aggregation is performed according to all predicted amino acid residue feature matrices in each category to generate an amino acid cluster feature matrix for each category, including:

[0014] Extracting the three-dimensional coordinates of the amino acid residues of each node according to the predicted amino acid residue feature matrix of each node;

[0015] DBSCAN clustering was performed based on the three-dimensional coordinates of the amino acid residues of each node, and the neighborhood radius of DBSCAN was set to 12 angstroms and the minimum number of points was set to 6;

[0016] Based on the spatial attention mechanism, the amino acid residues located in the tyrosine kinase region are weighted, and the sum of all predicted amino acid residue feature matrices under each category is calculated to obtain the amino acid cluster feature matrix of each category.

[0017] Furthermore, the amino acid cluster combination threshold is determined based on the predicted amino acid cluster feature matrix, the predicted amino acid cluster feature matrix is ​​hierarchically clustered using the amino acid cluster combination threshold, and all predicted amino acid cluster feature matrices in each category are weighted aggregated to generate multiple protein subunit feature matrices, including:

[0018] According to the predicted amino acid cluster feature matrix, the maximum distance between the centroid coordinates of all amino acid clusters is calculated, and 12% of the maximum distance is used as the amino acid cluster combination threshold;

[0019] Hierarchical clustering is performed on the amino acid clusters of each node according to the amino acid cluster combination threshold and the amino acid cluster centroid coordinates, and each category contains a plurality of predicted amino acid cluster feature matrices whose distances between the amino acid cluster centroid coordinates are less than the amino acid cluster combination threshold;

[0020] Based on the spatial attention mechanism, all predicted amino acid cluster feature matrices under each category are weighted, and the sum of all predicted amino acid cluster features of each category is calculated to obtain the protein subunit feature matrix of each category.

[0021] Furthermore, the extraction of amino acid residue features based on the three-dimensional spatial conformation of the composite mutant amino acid sequence containing multiple mutation sites includes:

[0022] Obtain predicted highly pathogenic EGFR single-point mutation sites, combine highly pathogenic EGFR mutation sites with sensitive mutation sites, and obtain a predicted set of compound mutation amino acid sequences;

[0023] Performing three-dimensional structural transformation on each compound mutation amino acid sequence in the compound mutation amino acid sequence prediction set to obtain a three-dimensional spatial conformation of the compound mutation amino acid sequence;

[0024] Traversing the three-dimensional spatial conformation of the compound mutant amino acid sequence, extracting all amino acid residue sequence characteristics, amino acid residue type characteristics and three-dimensional coordinates of all atoms in the compound mutant amino acid sequence;

[0025] Determining the key residue features of each amino acid residue based on the amino acid residue type features and the amino acid residue sequence features of each amino acid residue;

[0026] Determine the centroid coordinate characteristics of each amino acid residue according to the mean of the three-dimensional coordinates of all atoms of each amino acid residue;

[0027] The dihedral characteristics of each amino acid residue are determined based on the three-dimensional coordinates of all atoms in each amino acid residue.

[0028] Furthermore, the method further comprises:

[0029] The key residue features of each amino acid residue feature are identified, and the amino acid residue features of each node are weighted using an attention mechanism in the convolutional layer of the first graph neural network according to the key residue features.

[0030] Furthermore, the pathogenicity assessment of the predicted protein structure feature vector includes:

[0031] The predicted protein structure feature vectors are clustered, and the pathogenicity and drug sensitivity of unknown mutation sites in the same category as the known mutation sites are evaluated using the protein structure features containing the known mutation sites.

[0032] Furthermore, the method further comprises:

[0033] The predicted amino acid residue feature matrix is ​​clustered using the key amino acid residue features in the predicted amino acid residue features, and the pathogenicity, pathogenic mechanism and drug sensitivity of unknown mutation sites under the same category are evaluated using the predicted amino acid residue features containing known mutation sites.

[0034] In a second aspect, an embodiment of the present invention further provides an epidermal growth factor receptor pathogenicity prediction device, comprising:

[0035] Clinical mutation intersection analysis module, used to extract amino acid residue features based on the three-dimensional spatial conformation of compound mutation amino acid sequences;

[0036] The point-surface conformation flexible fusion module is used to predict protein structural features based on the characteristics of amino acid residues;

[0037] The point-surface conformation flexible fusion module comprises a first prediction unit for predicting amino acid residue features using a convolutional layer of a first graph neural network;

[0038] An amino acid cluster feature generating unit, used for generating amino acid cluster features according to predicted amino acid residue features;

[0039] A second prediction unit, used to predict the amino acid cluster features using a convolutional layer of a second graph neural network;

[0040] A protein subunit feature generation unit, used to generate protein subunit features based on predicted amino acid cluster features;

[0041] A third graph convolution unit, for predicting protein subunit features using the convolution layer of the third graph neural network;

[0042] A protein structure feature generating unit, used for generating protein structure features according to protein subunit features;

[0043] The point-surface conformation clustering dynamic evaluation module is used to evaluate the pathogenicity of mutation sites based on protein structural characteristics.

[0044] In a third aspect, an embodiment of the present invention further provides a server, including:

[0045] one or more processors;

[0046] a storage device for storing one or more programs,

[0047] When the one or more programs are executed by the one or more processors, the one or more processors implement the epidermal growth factor receptor pathogenicity prediction method provided in the above embodiment.

[0048] In a fourth aspect, an embodiment of the present invention further provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, are used to execute the method for predicting pathogenicity of epidermal growth factor receptor as provided in the above embodiment.

[0049] The embodiments of the present invention provide a method, device, server and medium for predicting the pathogenicity of epidermal growth factor receptor. The methods extract amino acid residue characteristics through the three-dimensional spatial conformation of the composite mutant amino acid sequence, predict the amino acid residue characteristics using a graph neural network, and aggregate the influence of mutation sites at the amino acid residue level; the amino acid residues are then aggregated according to the principle of spatial proximity to form higher-level amino acid clusters, and the amino acid clusters are predicted using a graph neural network to aggregate the influence of mutations into a higher-level amino acid cluster structure; the distances between the amino acid clusters are then used to aggregate the amino acid clusters into higher-level protein structural units - protein subunits, and the protein subunits are predicted again using a graph neural network, and the prediction results are mapped into feature vectors of the complete protein structure to classify the protein structure. It can predict the impact of mutations at different protein structure levels, gradually aggregate the impact of mutation sites into the entire protein structure, more accurately simulate the impact of mutations on the overall protein structure, improve the accuracy and comprehensiveness of the prediction, and use the classified protein structure to evaluate the pathogenicity and drug sensitivity of mutation sites, providing an important reference for the study of the biological characteristics of rare or unknown mutations, and providing a more scientific and reliable basis for clinical treatment, which is expected to bring more personalized treatment plans and better treatment effects to patients with EGFR mutant cancers. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] The accompanying drawings constituting a part of the present invention are used to provide a further understanding of the present invention. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the accompanying drawings:

[0051] Figure 1This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to Example 1 of the present invention;

[0052] Figure 2 This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to the second embodiment of the present invention;

[0053] Figure 3 This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to embodiment 3 of the present invention;

[0054] Figure 4 This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to Embodiment 4 of the present invention;

[0055] Figure 5 This is a schematic diagram of the structure of an epidermal growth factor receptor pathogenicity prediction device according to Example 6 of the present invention;

[0056] Figure 6 This is a schematic diagram of the structure of a server described in Embodiment 7 of the present invention. DETAILED DESCRIPTION

[0057] The present invention will be further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the present invention, rather than to limit the present invention. It should also be noted that, for ease of description, only parts related to the present invention, rather than all structures, are shown in the accompanying drawings.

[0058] Embodiment 1

[0059] Figure 1 This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to Example 1 of the present invention. This embodiment predicts the effect of mutation sites on protein structure through a graph neural network, and gradually aggregates from the amino acid residue level to the protein sub-level, and finally forms a protein structure feature vector for evaluating the pathogenicity of the mutation site. Specifically, the method includes the following steps:

[0060] S101, extracting amino acid residue features based on the three-dimensional spatial conformation of a composite mutant amino acid sequence containing multiple mutation sites, wherein the amino acid residue features include: amino acid residue type features, amino acid residue centroid coordinate features, key residue features and dihedral angle features, and the amino acid residue features are in matrix form.

[0061] When performing statistical analysis on mutation information, a specific mutation pattern was found. Compound mutations tend to superimpose on the two sensitive mutation sites of L858R and Del19 (exon 19 deletion), forming a compound mutation amino acid sequence including multiple mutation sites. An amino acid sequence is a sequence that represents the order of arrangement of multiple amino acids. When a compound mutation occurs, the amino acids at the corresponding mutation sites in the sequence will change. The three-dimensional conformation conversion software can be used to generate a three-dimensional conformation of the amino acid sequence, that is, the three-dimensional structure of the amino acid sequence. The three-dimensional conformation includes the three-dimensional structure of all amino acid sequences. By traversing the three-dimensional conformation using the tool, the spatial structural features of all amino acid residues that make up the three-dimensional conformation can be extracted. The type of amino acid residue can be identified from the spatial structural features, as well as whether the position of the amino acid residue is a key residue. The three-dimensional coordinates of all atoms of each amino acid residue in the spatial structural features can be used to obtain the center of mass coordinates and dihedral angles of the amino acid residue. These features can jointly describe the characteristics of the amino acid residue and form the characteristics of the amino acid residue.

[0062] For the convenience of calculation, the amino acid residue features are expressed in a matrix form, including the amino acid residue type features representing the amino acid residue type, the amino acid residue center of mass coordinate features representing the amino acid center of mass coordinates, the key residue features representing whether the amino acid residue is a key residue, and the dihedral angle features representing the conformation of the amino acid residue.

[0063] S102, generating an initial amino acid residue structure graph according to the amino acid residue characteristics, wherein the characteristics of each node in the initial amino acid residue structure graph are the corresponding amino acid residue characteristics; inputting the initial amino acid residue structure graph into a trained first graph neural network, and using the convolutional layer of the first graph neural network to update the amino acid residue characteristics of each node, to obtain a predicted amino acid residue feature matrix for each node.

[0064] The amino acid residues are arranged according to the amino acid sequence. In the three-dimensional spatial conformation generated by the amino acid sequence, the amino acid residues have a specific positional relationship. Based on the positional relationship between the amino acid residues, the initial amino acid residue structure graph is generated with the amino acid residues as nodes. The node positions in the graph correspond to the positions of the amino acid residues, and the edges in the graph are the interactions between the amino acid residues. The initial amino acid residue structure graph is updated using the first graph neural network, and the convolution layer is used to aggregate the neighbor nodes of each node. The formula is as follows:

[0065]

[0066] in, is the updated node feature matrix, σ is the activation function, such as ReLU, N(i) is the set of neighbor nodes of node i, cij is a normalization constant, which can be the edge feature e ij function, or it can be a constant 1, W (0) and b (0) is the learnable weight matrix and bias vector, x j is the node feature matrix of the neighboring nodes of node i.

[0067] The updated nodes contain the features of neighboring nodes, which reflects the changes in amino acid residues after being affected by the interacting adjacent amino acid residues, and obtains the predicted amino acid residue feature matrix of each node. When updating nodes, the features of mutated amino acid residues can be aggregated into the features of adjacent amino acid residues, reflecting the impact of the mutation site on other adjacent sites.

[0068] S103, according to the predicted amino acid residue feature matrix of each node, DBSCAN clustering is performed on the amino acid residues based on spatial proximity, and all predicted amino acid residue feature matrices in each category are weightedly aggregated to generate an amino acid cluster feature matrix for each category.

[0069] Multiple amino acid residues with spatial positional relationships often have a common characteristic. The predicted amino acid residue feature matrix containing the influence of mutation sites is classified based on spatial proximity. According to the amino acid residue centroid coordinate features in the predicted amino acid residue features, the DBSCAN clustering algorithm is used to classify the amino acid residue centroid coordinate features with spatial proximity, and then the predicted amino acid residue features are classified. All amino acid residues under each category are aggregated into amino acid clusters, which contain mutated amino acid residues or amino acid residues affected by mutations. When the predicted amino acid residue features under each category are aggregated, since the tyrosine kinase region is the key hub of EGFR signal transduction and the key position for tyrosine kinase inhibitor (TKI) targeting, the three-dimensional structure of amino acid residues here has a key influence on the biological activity of EGFR. It is necessary to emphasize the structural characteristics of amino acid residues located in the tyrosine kinase region from 682 to 958. Therefore, the weight of amino acid residues located in the tyrosine kinase region from 682 to 958 is increased through weighted operations to improve the influence of amino acid residues and generate the aggregated amino acid cluster feature matrix.

[0070] S104, generating an initial amino acid cluster structure diagram according to the amino acid cluster feature matrix, wherein the feature of each node in the initial amino acid cluster structure diagram is the corresponding amino acid cluster feature matrix; inputting the initial amino acid cluster structure diagram into the trained second graph neural network, and using the convolution layer of the second graph neural network to update the amino acid cluster feature matrix of each node to obtain a predicted amino acid cluster feature matrix of each node.

[0071] Different amino acid clusters formed by the polymerization of multiple spatially close amino acid residues show different characteristics, and each cluster is affected differently by mutations. The amino acid clusters formed after polymerization also have a relative spatial position relationship. According to the positional relationship between amino acid clusters, an initial amino acid cluster structure diagram is generated with amino acid clusters as nodes. The node positions in the diagram correspond to the positions of the amino acid clusters, and the edges in the diagram are the interactions between amino acid clusters. The initial amino acid cluster structure diagram is updated using the second graph neural network, and the convolutional layer is used to aggregate neighbor nodes for each node. The calculation formula is as follows:

[0072]

[0073] in, represents the predicted amino acid cluster feature matrix, σ is the activation function, such as ReLU, N cluster(k) represents the set of neighbor nodes of node k, c km is the normalization constant, W (1) and b (1) are the learnable weight matrices and bias vectors, is the node feature matrix of the neighboring nodes of node k.

[0074] The updated nodes contain the features of neighboring nodes, which can reflect the changes that occur when amino acid clusters are affected by neighboring amino acid clusters that interact with each other, and obtain the predicted amino acid cluster feature matrix of each node. When updating nodes, the features of multiple amino acid clusters affected by mutations can be aggregated into the features of adjacent amino acid clusters, reflecting the impact of amino acid clusters containing mutations on other adjacent amino acid clusters. By aggregating amino acid residues affected by mutations to form amino acid clusters affected by mutations, the impact of mutations in a larger range can be further realistically simulated through the impact between clusters.

[0075] S105, determining an amino acid cluster combination threshold according to the predicted amino acid cluster feature matrix, performing hierarchical clustering on the predicted amino acid cluster feature matrix using the amino acid cluster combination threshold, performing weighted aggregation on all predicted amino acid cluster feature matrices in each category, and generating multiple protein subunit feature matrices.

[0076] In order to consider the impact of mutations on protein structures at different levels, the centroid coordinates of amino acid clusters in the predicted amino acid cluster feature matrix were extracted, and hierarchical clustering was performed using the centroid coordinates of amino acid clusters. The distance threshold was set to 12% of the maximum value of the distance between the centroid coordinates of predicted amino acid clusters, and the predicted amino acid cluster feature matrix was classified. All amino acid clusters contained in each category were aggregated to form a larger structural unit, the protein subunit. Different weights were assigned to the amino acid clusters according to the distance between the amino acid clusters, and all predicted amino acid cluster feature matrices in each category were aggregated into a protein subunit feature matrix using different weights.

[0077] S106, generating an initial protein subunit structure diagram according to the protein subunit feature matrix, wherein the feature of each node in the initial protein subunit structure diagram is the corresponding protein subunit feature matrix; inputting the initial protein subunit structure diagram into the trained third graph neural network, and using the convolutional layer of the third graph neural network to update the protein subunit features of each node, to obtain a predicted protein subunit feature matrix of each node.

[0078] In order to comprehensively consider the impact of mutations on higher-level protein structures, the initial protein subunit structure graph is generated with protein subunits as nodes according to the positional relationship between protein subunits after aggregation. The node positions in the graph correspond to the positions of protein subunits, and the edges in the graph are the interactions between protein subunits. The initial protein subunit structure graph is updated using the third graph neural network, and the convolution layer is used to aggregate neighboring nodes for each node. The calculation formula is as follows:

[0079]

[0080] in, represents the predicted protein subunit feature matrix, σ is the activation function, such as ReLU, N Domain (n) represents the set of neighbor nodes, d np is the normalization constant, W (2) and b (2) are the learnable weight matrices and bias vectors, is the node feature matrix of the neighboring nodes of node n.

[0081] The updated nodes contain the features of neighboring nodes, and the predicted protein subunit feature matrix of each node is obtained. When updating nodes, the protein subunit features affected by the mutation can be aggregated into the adjacent protein subunit features, which can reflect the higher-level impact of the mutation on the overall protein structure.

[0082] S107, feature mapping is performed on all predicted protein subunit feature matrices to generate predicted protein structure feature vectors, and the predicted protein structure feature vectors are used to evaluate the pathogenicity of the mutation sites.

[0083] The predicted protein subunit features are mean pooled to obtain the global feature vector g, and then the fully connected layer is used for feature mapping to form the predicted protein structure feature vector y that expresses the complete protein structure characteristics. The formula is as follows:

[0084] y=σ(W (3) g+b (3) )

[0085] Among them, W (3) is the learnable weight matrix of the fully connected layer, b (3) is the bias vector of the fully connected layer, and σ is the sigmoid activation function.

[0086] The predicted protein structure feature vectors are clustered, and the protein structures in the same category have similarities. The predicted protein structure containing known mutation sites is identified by predicting the protein structure feature vectors, and the pathogenicity of the predicted protein structure of the same category is evaluated based on the characteristics of the known mutations. Proteins with the same or similar structures often show consistency or high similarity in their configurations. This structural homology makes them tend to be similar in biological properties such as drug sensitivity and pathogenicity. Therefore, the known mutation site information can be used to effectively predict and evaluate the biological properties of unknown or rare mutation sites of the same category.

[0087] In this embodiment, the first graph neural network is used to update and predict the amino acid residues to simulate the effect of mutations on the amino acid residues; the predicted amino acid residue feature matrix is ​​aggregated into an amino acid cluster feature matrix according to spatial proximity, and the amino acid clusters are updated and predicted using the second graph neural network, and the amino acid residues affected by the mutation are aggregated into amino acid clusters to simulate the effect of mutations from a higher level; the predicted amino acid cluster feature matrix affected by the mutation is then aggregated into a higher-level protein subunit feature matrix, and the protein subunits are predicted and updated using the third graph neural network to more accurately simulate the effect of mutations on different levels of the protein structure, and finally the predicted protein subunit feature matrix is ​​feature integrated to form a complete protein feature vector, and the complete protein feature vector is used for classification, and the known mutation sites of the same type are used as references to evaluate the pathogenicity and drug sensitivity of the unknown mutation sites of the same type, which can more comprehensively consider the effect of mutations on adjacent structures in different levels of protein structures, and the classification of mutations is more accurate, which can serve as an important reference when studying the pathogenicity and drug sensitivity of unknown mutations.

[0088] An optional implementation of this embodiment is that, in order to enhance the importance of key residues, when the convolutional layer of the first graph neural network is updated, the attention of key residues is enhanced by weighting. By identifying the key residue features of each amino acid residue feature, the amino acid residue features of each node are weighted using the attention mechanism in the convolutional layer of the first graph neural network according to the key residue features.

[0089] Key residues are amino acid residues located in the key functional region of the EGFR kinase structure. Changes (mutations) in these amino acid residues may greatly affect the function of the kinase and are closely related to the occurrence and development of the disease. The key residue features in the amino acid residue features are used to identify whether the amino acid residue is a key residue. When at least one of the node i or the neighbor node j is a key residue, an attention mechanism is added to the nodes representing the two amino acid residues for weighting. The attention weight calculation formula for the two nodes is as follows:

[0090]

[0091] Among them, exp represents the exponential function, LeakyReLU represents the activation function, a is the learnable attention vector, and a T is the transpose of vector a, ∥ represents vector concatenation, W (0) represents the learnable weight matrix, K represents the set of key amino acid residues, and N(i) represents the set of neighboring nodes of node i. if i∈K or j∈K means that at least one of node i or neighbor node j is a key residue, 0, otherwise represents other cases, that is, by judging whether node i or neighbor node j is a key residue, otherwise the node weight a ij is 0.

[0092] Correspondingly, the update formula of the convolutional layer of the first graph neural network is:

[0093]

[0094] By weighting key amino acid residues through the attention mechanism, the impact of mutations on amino acid residues can be simulated more realistically. Especially when key amino acid residues change, the impact of mutations can be predicted more accurately.

[0095] Embodiment 2

[0096] Figure 2This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to Example 2 of the present invention. This embodiment is optimized based on the above embodiment. In this embodiment, according to the predicted amino acid residue feature matrix of each node, DBSCAN clustering is performed on the amino acid residues based on spatial proximity, and all predicted amino acid residue feature matrices in each category are weighted aggregated to generate an amino acid cluster feature matrix for each category. The specific optimization is as follows:

[0097] Extracting the three-dimensional coordinates of the amino acid residues of each node according to the predicted amino acid residue feature matrix of each node;

[0098] DBSCAN clustering was performed based on the three-dimensional coordinates of the amino acid residues of each node, and the neighborhood radius of DBSCAN was set to 12 angstroms and the minimum number of points was set to 6;

[0099] Based on the spatial attention mechanism, the amino acid residues located in the tyrosine kinase region are weighted, and the sum of all predicted amino acid residue feature matrices under each category is calculated to obtain the amino acid cluster feature matrix of each category.

[0100] Accordingly, the method for predicting pathogenicity of epidermal growth factor receptor provided in this embodiment specifically includes:

[0101] S201, extracting amino acid residue features based on the three-dimensional spatial conformation of a composite mutant amino acid sequence containing multiple mutation sites, wherein the amino acid residue features include: amino acid residue type features, amino acid residue centroid coordinate features, key residue features and dihedral angle features, and the amino acid residue features are in matrix form.

[0102] S202, generating an initial amino acid residue structure graph according to the amino acid residue characteristics, wherein the characteristics of each node in the initial amino acid residue structure graph are the corresponding amino acid residue characteristics; inputting the initial amino acid residue structure graph into a trained first graph neural network, and using the convolutional layer of the first graph neural network to update the amino acid residue characteristics of each node, to obtain a predicted amino acid residue feature matrix for each node.

[0103] S203, extracting the three-dimensional coordinates of the amino acid residues of each node according to the predicted amino acid residue feature matrix of each node.

[0104] The predicted amino acid residue feature matrix includes amino acid residue centroid coordinate features after aggregating and updating the amino acid residue features of each node, and the three-dimensional coordinates of the amino acid centroid are determined according to the amino acid residue centroid coordinate features.

[0105] S204, performing DBSCAN clustering according to the three-dimensional coordinates of the amino acid residues of each node, setting the neighborhood radius of DBSCAN to 12 angstroms and the minimum number of points to 6.

[0106] The distance between the mass centers of any two amino acid residues was calculated using the Euclidean formula to obtain the amino acid residue distance matrix. The calculation formula is as follows:

[0107]

[0108] Where distance is the distance between the mass centers of two amino acid residues, (X i , Y i , Z i ) and (X j , Y j , Z j ) are the centroid coordinates of the two amino acid residues.

[0109] The DBSCAN algorithm is used to cluster the amino acid residues in the protein structure. By constructing a distance matrix between amino acid residues, the neighborhood radius (eps) is set to The minimum number of points (min_samples) was 6 to identify and classify groups of amino acid residues that are close to each other and dense in space.

[0110] S205, based on the spatial attention mechanism, weight the amino acid residues located in the tyrosine kinase region, and calculate the sum of all predicted amino acid residue feature matrices under each category to obtain the amino acid cluster feature matrix of each category.

[0111] Aggregate all amino acid residue features under each category into an amino acid cluster feature c k , calculate the spatial attention weights between any two nodes i and j in each category respectively The calculation formula is as follows:

[0112]

[0113] Among them, N spatial (k) represents the node set (i.e., amino acid residue set) under each category, exp represents the exponential function, and LeakyReLU scaled,l is the activation function, a is the learnable attention vector, and a T is the transpose of vector a, ∥ represents vector concatenation, W (1) represents the learnable weight matrix, and Represent the node feature matrices of nodes i and j respectively, Represents the node characteristics of all neighbor nodes.

[0114] According to the sequence characteristics of amino acid residues, the node features of the tyrosine kinase region located at positions 682 to 958 are scaled using the LeakyReLU function, and the scaling factor β (β>1) is defined to increase the node weight of the amino acid residue features in the tyrosine kinase region and emphasize the importance of these amino acid residues in the amino acid cluster:

[0115]

[0116] Among them, LeakyReLU(x)=max(0,x)+α·min(0,x) is the standard LeakyReLU activation function, α is a positive number less than 1, usually 0.01, which is used to control the slope when negative value input.

[0117] According to the interaction between the two nodes corresponding to the distance between the two nodes, weights are assigned to the nodes under each category. Multiple amino acid residues may present new characteristics after being combined into clusters. The tyrosine kinase region is the key site for the action of TKI targeted drugs. The structure of amino acid residues has a significant effect on the affinity of the drug. It is necessary to increase the weight of amino acid residues in the tyrosine kinase region during aggregation. Using spatial attention weights To aggregate all node feature vectors in amino acid cluster k Get the characteristic matrix of amino acid clusters

[0118]

[0119] in, is the total attention weight of node j in cluster k.

[0120] S206, generating an initial amino acid cluster structure diagram according to the amino acid cluster feature matrix, wherein the feature of each node in the initial amino acid cluster structure diagram is the corresponding amino acid cluster feature matrix; inputting the initial amino acid cluster structure diagram into the trained second graph neural network, and using the convolution layer of the second graph neural network to update the amino acid cluster feature matrix of each node, to obtain a predicted amino acid cluster feature matrix of each node.

[0121] S207, determining an amino acid cluster combination threshold according to the predicted amino acid cluster feature matrix, performing hierarchical clustering on the predicted amino acid cluster feature matrix using the amino acid cluster combination threshold, performing weighted aggregation on all predicted amino acid cluster feature matrices in each category, and generating multiple protein subunit feature matrices.

[0122] S208, generating an initial protein subunit structure diagram according to the protein subunit feature matrix, wherein the feature of each node in the initial protein subunit structure diagram is the corresponding protein subunit feature matrix; inputting the initial protein subunit structure diagram into the trained third graph neural network, and using the convolutional layer of the third graph neural network to update the protein subunit features of each node, to obtain a predicted protein subunit feature matrix for each node.

[0123] S209, feature mapping is performed on all predicted protein subunit feature matrices to generate predicted protein structure feature vectors, and the predicted protein structure feature vectors are used to evaluate the pathogenicity of the mutation sites.

[0124] In this embodiment, the updated predicted amino acid residue feature matrix is ​​classified by DBSCAN based on spatial proximity, all amino acid residues in the same category are aggregated into amino acid clusters, and the amino acid residues affected by the mutation are used to form a higher protein hierarchical structure, and the impact of the mutation is predicted from a higher protein structure level. At the same time, the spatial attention mechanism is used to weight the amino acid residues in the tyrosine kinase region, and the importance of the amino acid residues in the tyrosine kinase region is emphasized when the amino acid residues are aggregated into amino acid clusters, which can more realistically simulate the characteristics of the amino acid clusters affected by the mutation.

[0125] Embodiment 3

[0126] Figure 3 This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to Example 3 of the present invention. This embodiment is optimized based on the above embodiment. In this embodiment, the amino acid cluster combination threshold is determined according to the predicted amino acid cluster feature matrix, and the predicted amino acid cluster feature matrix is ​​hierarchically clustered using the amino acid cluster combination threshold. All predicted amino acid cluster feature matrices in each category are weighted aggregated to generate multiple protein subunit feature matrices. The specific optimization is as follows:

[0127] According to the predicted amino acid cluster feature matrix, the maximum distance between the centroid coordinates of all amino acid clusters is calculated, and 12% of the maximum distance is used as the amino acid cluster combination threshold;

[0128] Hierarchical clustering is performed on the amino acid clusters of each node according to the amino acid cluster combination threshold and the amino acid cluster centroid coordinates, and each category contains a plurality of predicted amino acid cluster feature matrices whose distances between the amino acid cluster centroid coordinates are less than the amino acid cluster combination threshold;

[0129] Based on the spatial attention mechanism, all predicted amino acid cluster feature matrices under each category are weighted, and the sum of all predicted amino acid cluster features of each category is calculated to obtain the protein subunit feature matrix of each category.

[0130] Accordingly, the method for predicting pathogenicity of epidermal growth factor receptor provided in this embodiment specifically includes:

[0131] S301, extracting amino acid residue features based on the three-dimensional spatial conformation of a composite mutant amino acid sequence containing multiple mutation sites, wherein the amino acid residue features include: amino acid residue type features, amino acid residue centroid coordinate features, key residue features and dihedral angle features, and the amino acid residue features are in matrix form.

[0132] S302, generating an initial amino acid residue structure graph according to the amino acid residue characteristics, wherein the characteristics of each node in the initial amino acid residue structure graph are the corresponding amino acid residue characteristics; inputting the initial amino acid residue structure graph into the trained first graph neural network, and using the convolutional layer of the first graph neural network to update the amino acid residue characteristics of each node, to obtain a predicted amino acid residue feature matrix for each node.

[0133] S303, according to the predicted amino acid residue feature matrix of each node, DBSCAN clustering is performed on the amino acid residues based on spatial proximity, and all predicted amino acid residue feature matrices in each category are weightedly aggregated to generate an amino acid cluster feature matrix for each category.

[0134] S304, generating an initial amino acid cluster structure diagram according to the amino acid cluster feature matrix, wherein the feature of each node in the initial amino acid cluster structure diagram is the corresponding amino acid cluster feature matrix; inputting the initial amino acid cluster structure diagram into the trained second graph neural network, and using the convolution layer of the second graph neural network to update the amino acid cluster feature matrix of each node, to obtain a predicted amino acid cluster feature matrix of each node.

[0135] S305, calculating the maximum distance between the centroid coordinates of all amino acid clusters according to the predicted amino acid cluster feature matrix, and taking 12% of the maximum distance as the amino acid cluster combination threshold;

[0136] The centroid coordinates of the amino acid clusters were determined based on the centroid coordinate features of the amino acid clusters in the predicted amino acid cluster feature matrix, and the distance between the centroids of any two amino acid clusters was calculated using the Euclidean formula to obtain the amino acid cluster distance matrix. Taking 12% of the maximum distance between the centroids of any two amino acid clusters as the amino acid cluster combination threshold can help identify amino acid clusters with significant interactions.

[0137] S306, hierarchical clustering is performed on the amino acid clusters of each node according to the amino acid cluster combination threshold and the amino acid cluster centroid coordinates, and each category includes multiple predicted amino acid cluster feature matrices whose distances between the amino acid cluster centroid coordinates are less than the amino acid cluster combination threshold.

[0138] The amino acid cluster distance matrix was hierarchically clustered using an amino acid cluster combination threshold of 12% of the maximum distance between the amino acid cluster centroids, and the amino acid clusters under each category were combined into protein subunits.

[0139] S307, based on the spatial attention mechanism, weight all predicted amino acid cluster feature matrices under each category, and calculate the sum of all predicted amino acid cluster features of each category to obtain the protein subunit feature matrix of each category.

[0140] Aggregate all amino acid cluster features under each category into a protein subunit feature d k , respectively calculate the spatial attention weights between any two amino acid clusters m and n under each category The calculation formula is as follows:

[0141]

[0142] Where m,n∈N Domain (D k ), N Domain (D k ) represents the node set under each category, exp represents the exponential function, LeakyReLU is the activation function, a is the learnable attention vector, a T is the transpose of vector a, ∥ represents vector concatenation, W (2) represents the learnable weight matrix, and Represent the node feature matrices of nodes m and n respectively.

[0143] Using spatial attention weights Aggregate all amino acid cluster feature matrices under each category into a protein subunit feature matrix

[0144]

[0145] in, Indicates that the amino acid cluster m is in the structural unit D k The total attention weight within, N Domain (D k ) represents the clusters of amino acids that make up protein subunits.

[0146] S308, generating an initial protein subunit structure diagram according to the protein subunit feature matrix, wherein the feature of each node in the initial protein subunit structure diagram is the corresponding protein subunit feature matrix; inputting the initial protein subunit structure diagram into the trained third graph neural network, and using the convolutional layer of the third graph neural network to update the protein subunit features of each node, to obtain a predicted protein subunit feature matrix for each node.

[0147] S309, feature mapping is performed on all predicted protein subunit feature matrices to generate predicted protein structure feature vectors, and the predicted protein structure feature vectors are used to evaluate the pathogenicity of the mutation sites.

[0148] In this embodiment, the updated predicted amino acid cluster feature matrix is ​​classified by hierarchical clustering, all amino acid clusters under the same category are aggregated into protein subunits, and the affected amino acid clusters are used to form a higher protein hierarchical structure. At the same time, the amino acid clusters are weighted by the spatial attention mechanism to emphasize the different importance of amino acid clusters with different characteristics, predict the impact of mutations from a higher protein structure level, and more comprehensively express the impact of mutations on different levels of protein structure.

[0149] Embodiment 4

[0150] Figure 4 This is a flow chart of a method for predicting pathogenicity of epidermal growth factor receptor according to Example 4 of the present invention. This example is optimized based on the above example. In this example, the amino acid residue features are extracted based on the three-dimensional spatial conformation of the composite mutant amino acid sequence containing multiple mutation sites. The specific optimization is as follows:

[0151] Obtain predicted highly pathogenic EGFR single-point mutation sites, combine highly pathogenic EGFR mutation sites with sensitive mutation sites, and obtain a predicted set of compound mutation amino acid sequences;

[0152] Performing three-dimensional structural transformation on each compound mutation amino acid sequence in the compound mutation amino acid sequence prediction set to obtain a three-dimensional spatial conformation of the compound mutation amino acid sequence;

[0153] Traversing the three-dimensional spatial conformation of the compound mutant amino acid sequence, extracting all amino acid residue sequence characteristics, amino acid residue type characteristics and three-dimensional coordinates of all atoms in the compound mutant amino acid sequence;

[0154] Determining the key residue features of each amino acid residue based on the amino acid residue type features and the amino acid residue sequence features of each amino acid residue;

[0155] Determine the centroid coordinate characteristics of each amino acid residue according to the mean of the three-dimensional coordinates of all atoms of each amino acid residue;

[0156] The dihedral characteristics of each amino acid residue are determined based on the three-dimensional coordinates of all atoms in each amino acid residue.

[0157] Accordingly, the method for predicting pathogenicity of epidermal growth factor receptor provided in this embodiment specifically includes:

[0158] S401, obtain the predicted highly pathogenic EGFR single point mutation sites, combine the highly pathogenic EGFR mutation sites with the sensitive mutation sites, and obtain a predicted set of compound mutation amino acid sequences.

[0159] The AlphaMissense prediction model is used to predict all potential missense mutation sites of EGFR, and clinical data and EGFR mutation samples in public gene databases are obtained. The predicted potential missense mutation sites can be classified into known mutations and unknown mutations using the Venn diagram method and the acquired mutation sample data. There are relatively or relatively mature treatment methods for known mutation sites in clinical practice. The unknown mutation sites can be treated by treatment or medication by analyzing the correlation or similarity between the unknown mutation sites and the known mutation sites, referring to the treatment methods of the known mutation sites.

[0160] The predicted potential missense mutation sites include possible benign mutations and possible pathogenic mutations of key concern. The possible pathogenic mutations of key concern are classified using the SIFT model based on sequence conservation, the PROVEAN model based on machine learning, and the AlphaMissense model that integrates multiple principles. The prediction results of each model mark the mutation site as pathogenic or benign according to the cutoff value. Specifically, the cutoff value of the SIFT model is set to 0.05, and mutations below this value in this model are highly pathogenic; the cutoff value of the PROVEAN model is set to -2.5, and mutations below this value in this model are pathogenic; the cutoff value of the PolyPhen-2 model is set to 0.99, and values ​​close to 1 in this model indicate highly likely pathogenicity; the cutoff value of the AlphaMissense model is set to 0.99, and mutations above this value in this model are highly pathogenic. If the mutation site is predicted to be pathogenic in all models, it is classified as highly pathogenic. If three models predict it to be pathogenic, it is classified as relatively highly pathogenic. If two models predict it to be pathogenic, it is classified as moderately pathogenic. The rest are classified as low pathogenic. The highly pathogenic mutation sites predicted to be pathogenic by all models are selected and combined with two sensitive mutations, L858R and Del19 (exon 19 deletion), to obtain a composite mutation amino acid sequence prediction set consisting of multiple composite mutation amino acid sequences.

[0161] S402, performing three-dimensional structural transformation on each compound mutant amino acid sequence in the compound mutant amino acid sequence prediction set to obtain a three-dimensional spatial conformation of the compound mutant amino acid sequence.

[0162] In order to better simulate the impact of mutation sites on the overall protein structure, the three-dimensional spatial conformation conversion tool is used to convert the compound mutation amino acid sequence into the corresponding three-dimensional spatial conformation. Specifically, the AlphaFold3 tool can be used to convert the compound mutation amino acid sequence into a three-dimensional spatial conformation. The model uses an improved Transformer architecture and conditional diffusion model to predict and output the three-dimensional structure of the protein.

[0163] S403, traversing the three-dimensional spatial conformation of the compound mutant amino acid sequence, extracting all amino acid residue order characteristics, amino acid residue type characteristics and three-dimensional coordinates of all atoms in the compound mutant amino acid sequence.

[0164] The three-dimensional conformation of the compound mutant amino acid sequence was traversed using tools. Specifically, the PDBParser and PDBIO modules in the Python machine learning library Biopython were used to traverse all amino acid residues and each amino acid chain in the three-dimensional conformation (PDB file or CIF file) and extract the three-dimensional coordinates (x i ,y i ,z i ).

[0165] The type of amino acid residue can be determined by the atomic arrangement of each amino acid residue. According to the 20 types of amino acids, a 20-dimensional binary vector representing the type of amino acid residue is generated as the amino acid residue type feature using one-hot encoding. Different types of amino acid residues are distinguished by one of the 20 elements of the vector being 1 and the other elements being 0. For example, the encoding of alanine is [1,0,0,…,0], the encoding of cysteine ​​is [0,1,0,…,0], and so on. And when traversing, the order of amino acid residues can also be determined according to the order of traversal.

[0166] S404, determining the key residue feature of each amino acid residue according to the amino acid residue type feature and the amino acid residue sequence feature of each amino acid residue.

[0167] Some amino acid residues are located in the key functional region of the EGFR kinase structure. The amino acid residues in this region play a vital role in the catalytic reaction of the kinase. They are directly or indirectly involved in the catalytic reaction of the kinase, not only in the binding and hydrolysis of ATP, but also in the recognition and phosphorylation of the substrate, and the regulation of kinase activity. Changes (mutations) in these amino acid residues may affect the function of the kinase and are closely related to the occurrence and development of the disease. Exemplary, key amino acid residues include Lys.K745, Glu.E762, Met.M766, Lys.K777, Val.V834, His.H835, Arg.R836, Thr.T854, Asp.D855, Phe.F856, Gly.G857, Lys.K860, Val.V876, Val.V884, etc. A key amino acid residue is a specific type of amino acid residue at a specific position in an amino acid sequence. By traversing the obtained amino acid residue type characteristics and amino acid residue order characteristics, it is possible to identify whether the amino acid residue is a key residue and generate the corresponding amino acid residue characteristics. The amino acid residue characteristics can be represented by a one-dimensional binary element, for example: 0 represents a non-key amino acid residue, and 1 represents a key amino acid residue.

[0168] S405, determining the centroid coordinate characteristics of each amino acid residue according to the mean of the three-dimensional coordinates of all atoms of each amino acid residue.

[0169] The average value of all atomic coordinates of each amino acid residue is used to determine the centroid coordinates (Xc, Yc, Zc) of the amino acid residue, indicating the position of the amino acid residue in the three-dimensional conformation:

[0170] Xc=(x 1 +x 2 +...+x n ) / n

[0171] Yc=(y 1 +y 2 +...+y n ) / n

[0172] Zc=(z 1 +z 2 +...+z n ) / n

[0173] Where n is the total number of atoms in the residue, x i ,y i 、z i Respectively represent the coordinates of the ith atom in the X-axis, Y-axis and Z-axis directions. The center-of-mass coordinates of the amino acid residue are used to generate the amino acid residue center-of-mass coordinate features consisting of three elements.

[0174] S406, determining the dihedral angle characteristics of each amino acid residue based on the three-dimensional coordinates of all atoms of each amino acid residue.

[0175] The main chain dihedral angles phi (φ) and psi (ψ) are two important parameters used to describe the torsion angles of the amino acid main chain. These two angles define the relative position and direction of the amino acid residues in the protein chain. These angles are obtained by calculating the three-dimensional coordinates of five key atoms on the amino acid main chain, including the carbon atom C′ (the C atom of the previous amino acid), the amino nitrogen atom N, the central carbon atom Cα, the carboxyl carbon atom C and the nitrogen atom N″ (the N atom of the next amino acid) on the main chain.

[0176] Suppose the three-dimensional coordinates of the five atoms are C′(x1,y1,z1), N(x2,y2,z2), Cα(x3,y3,z3), C(x4,y4,z4), and N″(x5,y5,z5). Then the vectors between the atoms are expressed as:

[0177] Vector AB = NC′ = (x2-x1, y2-y1, z2-z1)

[0178] Vector BC = Cα-N = (x3-x2, y3-y2, z3-z2)

[0179] Vector CD = C-Cα = (x4-x3, y4-y3, z4-z3)

[0180] Vector DE = N″-C = (x5-x4, y5-y4, z5-z4)

[0181] Use the cross product to calculate the normal vector: n1 = AB × BC, n2 = BC × CD;

[0182] Calculate the cosine of the angle between vectors by dot product: phi =n1·n2; dot psi =n2·DE;

[0183] The norm of the vector is expressed as: norm_n1 = |n1|, norm_n2 = |n2|, norm_DE = |DE|;

[0184] From this we can see that the dihedral angle can be expressed as:

[0185] Use the arccos function of the NumPy library in Python to calculate the size of the dihedral angles:

[0186] phi=arccos(cos_phi), psi=arccos(cos_psi).

[0187] The size of the dihedral angle is used to generate a dihedral feature consisting of two elements.

[0188] The amino acid residue type features, amino acid residue centroid coordinate features, key residue features and dihedral angle features are spliced ​​to form amino acid residue features in matrix form.

[0189] S407, generating an initial amino acid residue structure graph according to the amino acid residue characteristics, wherein the characteristics of each node in the initial amino acid residue structure graph are the corresponding amino acid residue characteristics; inputting the initial amino acid residue structure graph into the trained first graph neural network, and using the convolutional layer of the first graph neural network to update the amino acid residue characteristics of each node, to obtain a predicted amino acid residue feature matrix for each node.

[0190] S408, according to the predicted amino acid residue feature matrix of each node, DBSCAN clustering is performed on the amino acid residues based on spatial proximity, and all predicted amino acid residue feature matrices in each category are weighted aggregated to generate an amino acid cluster feature matrix for each category.

[0191] S409, generating an initial amino acid cluster structure diagram according to the amino acid cluster feature matrix, wherein the feature of each node in the initial amino acid cluster structure diagram is the corresponding amino acid cluster feature matrix; inputting the initial amino acid cluster structure diagram into the trained second graph neural network, and using the convolutional layer of the second graph neural network to update the amino acid cluster feature matrix of each node, to obtain a predicted amino acid cluster feature matrix of each node.

[0192] S410, determining an amino acid cluster combination threshold according to the predicted amino acid cluster feature matrix, performing hierarchical clustering on the predicted amino acid cluster feature matrix using the amino acid cluster combination threshold, performing weighted aggregation on all predicted amino acid cluster feature matrices in each category, and generating multiple protein subunit feature matrices.

[0193] S411, generating an initial protein subunit structure diagram according to the protein subunit feature matrix, wherein the feature of each node in the initial protein subunit structure diagram is the corresponding protein subunit feature matrix; inputting the initial protein subunit structure diagram into the trained third graph neural network, and using the convolutional layer of the third graph neural network to update the protein subunit features of each node, to obtain a predicted protein subunit feature matrix for each node.

[0194] S412, feature mapping is performed on all predicted protein subunit feature matrices to generate predicted protein structure feature vectors, and the predicted protein structure feature vectors are used to evaluate the pathogenicity of the mutation sites.

[0195] This example predicts mutation sites by using clinical sample data and somatic mutation database data, obtains potential missense mutation sites including known mutations and unknown mutations, uses different pathogenicity analysis models to evaluate the pathogenicity of all potential missense mutations, and combines the highly pathogenic mutations obtained comprehensively with common sensitive mutation sites to form a composite mutation amino acid sequence; uses the composite mutation amino acid sequence to generate a three-dimensional spatial conformation, traverses the three-dimensional spatial conformation of the composite mutation amino acid sequence, and obtains the characteristics of all amino acid residues, which are used to predict and analyze the effects of multiple mutations on different levels of protein structure.

[0196] Embodiment 5

[0197] This embodiment is optimized based on the above embodiment. In this embodiment, the pathogenicity of the predicted protein structure feature vector is evaluated, and the specific optimization is as follows:

[0198] The predicted protein structure feature vectors are clustered, and the pathogenicity and drug sensitivity of unknown mutation sites in the same category as the known mutation sites are evaluated using the protein structure features containing the known mutation sites.

[0199] Hierarchical clustering method is used to perform cluster analysis on the predicted protein structure feature vectors, divide them into different categories, and identify the protein structure features containing known mutations in each category. Since the protein structure features in the same category are highly similar, the three-dimensional conformational change patterns after binding with the same drug are also consistent. The pathogenicity and drug sensitivity information of known mutations in clinical data can be referred to to evaluate the pathogenicity and drug sensitivity of protein structures with unknown mutations in the same category, providing scientific and precise guidance for the clinical treatment and drug matching of unknown mutation sites, and helping to achieve personalized drug matching.

[0200] Optionally, the predicted amino acid residue feature matrix is ​​clustered using key amino acid residue features in the predicted amino acid residue features, and the pathogenicity, pathogenic mechanism and drug sensitivity of unknown mutation sites in the same category are evaluated using predicted amino acid residue features containing known mutation sites.

[0201] By analyzing the situation in which the key residue features in the predicted amino acid residue features are affected by mutations, the predicted amino acid residue features in each category after classification have similar conformations and functions. Exemplarily, according to the predicted amino acid residue feature matrix, the amino acid residues are divided into nine categories: α, β, γ, δ, ε, ζ, η, θ, ι, and the three-dimensional conformational features of the tyrosine kinase region of the amino acid residues in each category are similar. Among them, the α-class mutants involve clinical classic drug-resistant mutations T790M, C797S, G719D, and S768I_G719C compound mutations, etc. According to known mutations, it can be known that the mutants under this classification may have drug-insensitive functions, so they can be used as a reference for pathogenicity; the kinase central domain conformation of the β-class mutants has a distance difference of less than 3.60 angstroms, which can form a salt bridge, Thr.T854, Asp.D855, Phe.F856, Gly.G857 The dihedral angle value of the amino acid residue determines the orientation of phenylalanine. At this time, aspartic acid can form a coordination bond with Mg-ATP. This conformational change provides the necessary spatial structure and chemical environment for the binding of ATP. According to the known mutations, it can be known that the mutants under this classification may have abnormal protein activation function and can be used as a reference for pathogenicity; by analogy, γ, δ, ε, ζ, and θ mutants can be used as references for potential pathogenicity and need to be paid attention to; the pathogenic function of η and ι mutants is still in doubt and needs further analysis and verification.

[0202] Embodiment 6

[0203] Figure 5 Schematic diagram of the structure of an epidermal growth factor receptor pathogenicity prediction device according to Example 6 of the present invention, as shown in Figure 5 As shown, the device comprises:

[0204] A clinical mutation intersection analysis module 810 is used to extract amino acid residue features based on the three-dimensional spatial conformation of the compound mutation amino acid sequence;

[0205] Point-surface conformation flexible fusion module 820, used to predict protein structural features based on amino acid residue features;

[0206] The point-surface conformation flexible fusion module 820 includes a first prediction unit 821 for predicting amino acid residue features using a convolutional layer of a first graph neural network;

[0207] An amino acid cluster feature generating unit 822, used to generate amino acid cluster features according to the predicted amino acid residue features;

[0208] A second prediction unit 823, used to predict the amino acid cluster features using the convolutional layer of the second graph neural network;

[0209] A protein subunit feature generating unit 824, used to generate protein subunit features according to the predicted amino acid cluster features;

[0210] A third graph convolution unit 825, for predicting protein subunit features using a convolutional layer of a third graph neural network;

[0211] A protein structure feature generating unit 826, used to generate protein structure features according to protein subunit features;

[0212] The point-surface conformation clustering dynamic evaluation module 830 is used to evaluate the pathogenicity of mutation sites based on protein structural features.

[0213] The epidermal growth factor receptor pathogenicity prediction system described in this embodiment extracts amino acid residue features, and predicts mutation sites from low-level structures to high-level structures according to the hierarchical structure of the protein, and uses clustering and spatial attention mechanisms to aggregate low-level structures into higher-level structures, ultimately forming complete protein structural features, classifying the complete protein structural features, and using the characteristics of known mutation sites in the same category to evaluate the pathogenicity and drug sensitivity of unknown mutation sites. It can more comprehensively consider the impact of mutations on adjacent structures in different hierarchical structures of proteins, and can serve as an important reference for the study of unknown mutations.

[0214] Based on the above embodiments, the amino acid cluster feature generation unit includes:

[0215] An amino acid residue coordinate extraction subunit, used to extract the three-dimensional coordinates of the amino acid residues of each node according to the predicted amino acid residue feature matrix of each node;

[0216] The DBSCAN clustering subunit is used to classify the predicted amino acid residue feature matrix of each node.

[0217] The tyrosine kinase region weighting subunit is used to weight the amino acid residues located in the tyrosine kinase region.

[0218] Based on the above embodiments, the protein subunit feature generation unit includes:

[0219] An amino acid cluster combination threshold calculation subunit, used to calculate the amino acid cluster combination threshold according to the maximum distance between the centroid coordinates of all amino acid clusters in the predicted amino acid cluster feature matrix;

[0220] A hierarchical clustering subunit is used to classify the predicted amino acid cluster feature matrix of each node;

[0221] The amino acid cluster weighting subunit is used to weight all predicted amino acid cluster feature matrices under each category.

[0222] The epidermal growth factor receptor pathogenicity prediction system provided in the embodiment of the present invention can execute the epidermal growth factor receptor pathogenicity prediction method provided in any embodiment of the present invention, and has the corresponding functional modules and beneficial effects of the execution method.

[0223] Embodiment 7

[0224] Figure 6 This is a schematic diagram of the structure of a server described in Embodiment 7 of the present invention. Figure 6 A block diagram of an exemplary server 12 suitable for use in implementing embodiments of the present invention is shown. Figure 6 The server 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0225] like Figure 6 As shown, the server 12 is represented in the form of a general-purpose computing device. The components of the server 12 may include, but are not limited to: one or more processors or processing units;

[0226] System memory, bus 18 that connects various system components including system memory 28 and processing unit 16.

[0227] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0228] The server 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the server 12, including volatile and non-volatile media, removable and non-removable media.

[0229] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The server 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 6 not shown, usually called a "hard drive"). Although Figure 6Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0230] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0231] The server 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), may communicate with one or more devices that enable a user to interact with the device / system / server 12, and / or may communicate with any device that enables the server 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed via an input / output (I / O) interface 22. In addition, the server 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 20. As shown, the network adapter 20 communicates with other modules of the server 12 via a bus 18. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the server 12, including, but not limited to, microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0232] The processing unit 16 executes various functional applications and data processing by running the programs stored in the system memory 28, such as implementing the epidermal growth factor receptor pathogenicity prediction method provided in the embodiment of the present invention.

[0233] Embodiment 8

[0234] Embodiment 8 of the present invention further provides a storage medium comprising computer executable instructions, which, when executed by a computer processor, are used to execute the method for predicting pathogenicity of epidermal growth factor receptor as provided in the above embodiments.

[0235] The computer storage medium of the embodiment of the present invention can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In this document, a computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0236] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0237] Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0238] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0239] Note that the above are only preferred embodiments of the present invention and the technical principles used. Those skilled in the art will understand that the present invention is not limited to the specific embodiments described herein, and that various obvious changes, readjustments and substitutions can be made by those skilled in the art without departing from the scope of protection of the present invention. Therefore, although the present invention has been described in more detail through the above embodiments, the present invention is not limited to the above embodiments, and may include more other equivalent embodiments without departing from the concept of the present invention, and the scope of the present invention is determined by the scope of the appended claims.

Claims

1. A method for predicting pathogenicity of epidermal growth factor receptor, characterized in that: include: Extracting amino acid residue features based on the three-dimensional spatial conformation of a composite mutant amino acid sequence containing multiple mutation sites, wherein the amino acid residue features include: amino acid residue type features, amino acid residue centroid coordinate features, key residue features and dihedral angle features, and the amino acid residue features are in matrix form; According to the amino acid residue characteristics, an initial amino acid residue structure graph is generated, wherein the characteristics of each node in the initial amino acid residue structure graph are the corresponding amino acid residue characteristics; the initial amino acid residue structure graph is input into the trained first graph neural network, and the amino acid residue characteristics of each node are updated using the convolution layer of the first graph neural network to obtain a predicted amino acid residue feature matrix for each node; According to the predicted amino acid residue feature matrix of each node, DBSCAN clustering is performed on the amino acid residues based on spatial proximity, and all predicted amino acid residue feature matrices in each category are weighted aggregated to generate the amino acid cluster feature matrix of each category; According to the amino acid cluster feature matrix, an initial amino acid cluster structure diagram is generated, wherein the feature of each node in the initial amino acid cluster structure diagram is the corresponding amino acid cluster feature matrix; the initial amino acid cluster structure diagram is input into the trained second graph neural network, and the amino acid cluster feature matrix of each node is updated by using the convolution layer of the second graph neural network to obtain a predicted amino acid cluster feature matrix of each node; According to the predicted amino acid cluster feature matrix, the amino acid cluster combination threshold is determined, the predicted amino acid cluster feature matrix is ​​hierarchically clustered using the amino acid cluster combination threshold, all predicted amino acid cluster feature matrices in each category are weightedly aggregated to generate multiple protein subunit feature matrices; According to the protein subunit feature matrix, an initial protein subunit structure graph is generated, wherein the feature of each node in the initial protein subunit structure graph is the corresponding protein subunit feature matrix; the initial protein subunit structure graph is input into the trained third graph neural network, and the protein subunit feature of each node is updated by using the convolution layer of the third graph neural network to obtain a predicted protein subunit feature matrix of each node; All predicted protein subunit feature matrices are feature mapped to generate predicted protein structure feature vectors, which are then used to evaluate the pathogenicity of mutation sites.

2. The method according to claim 1, characterized in that According to the predicted amino acid residue feature matrix of each node, DBSCAN clustering is performed on the amino acid residues based on spatial proximity, and all predicted amino acid residue feature matrices in each category are weighted aggregated to generate an amino acid cluster feature matrix for each category, including: Extracting the three-dimensional coordinates of the amino acid residues of each node according to the predicted amino acid residue feature matrix of each node; DBSCAN clustering was performed based on the three-dimensional coordinates of the amino acid residues of each node, and the neighborhood radius of DBSCAN was set to 12 angstroms and the minimum number of points was set to 6; Based on the spatial attention mechanism, the amino acid residues located in the tyrosine kinase region are weighted, and the sum of all predicted amino acid residue feature matrices under each category is calculated to obtain the amino acid cluster feature matrix of each category.

3. The method according to claim 1, characterized in that: The method comprises: determining an amino acid cluster combination threshold according to the predicted amino acid cluster feature matrix, performing hierarchical clustering on the predicted amino acid cluster feature matrix using the amino acid cluster combination threshold, performing weighted aggregation on all predicted amino acid cluster feature matrices in each category, and generating multiple protein subunit feature matrices, including: According to the predicted amino acid cluster feature matrix, the maximum distance between the centroid coordinates of all amino acid clusters is calculated, and 12% of the maximum distance is used as the amino acid cluster combination threshold; Hierarchical clustering is performed on the amino acid clusters of each node according to the amino acid cluster combination threshold and the amino acid cluster centroid coordinates, and each category contains a plurality of predicted amino acid cluster feature matrices whose distances between the amino acid cluster centroid coordinates are less than the amino acid cluster combination threshold; Based on the spatial attention mechanism, all predicted amino acid cluster feature matrices under each category are weighted, and the sum of all predicted amino acid cluster features of each category is calculated to obtain the protein subunit feature matrix of each category.

4. The method according to claim 1, characterized in that The method of extracting amino acid residue features based on the three-dimensional spatial conformation of a composite mutant amino acid sequence containing multiple mutation sites includes: Obtain predicted highly pathogenic EGFR single-point mutation sites, combine highly pathogenic EGFR mutation sites with sensitive mutation sites, and obtain a predicted set of compound mutation amino acid sequences; Performing three-dimensional structural transformation on each compound mutation amino acid sequence in the compound mutation amino acid sequence prediction set to obtain a three-dimensional spatial conformation of the compound mutation amino acid sequence; Traversing the three-dimensional spatial conformation of the compound mutant amino acid sequence, extracting all amino acid residue sequence characteristics, amino acid residue type characteristics and three-dimensional coordinates of all atoms in the compound mutant amino acid sequence; Determining the key residue features of each amino acid residue based on the amino acid residue type features and the amino acid residue sequence features of each amino acid residue; Determine the centroid coordinate characteristics of each amino acid residue according to the mean of the three-dimensional coordinates of all atoms of each amino acid residue; The dihedral characteristics of each amino acid residue are determined based on the three-dimensional coordinates of all atoms in each amino acid residue.

5. The method according to claim 1, characterized in that The method further comprises: The key residue features of each amino acid residue feature are identified, and the amino acid residue features of each node are weighted using an attention mechanism in the convolutional layer of the first graph neural network according to the key residue features.

6. The method according to claim 1, characterized in that The pathogenicity assessment of the predicted protein structure feature vector includes: The predicted protein structure feature vectors are clustered, and the pathogenicity and drug sensitivity of unknown mutation sites in the same category as the known mutation sites are evaluated using the protein structure features containing the known mutation sites.

7. The method according to claim 1, characterized in that The method further comprises: The predicted amino acid residue feature matrix is ​​clustered using the key amino acid residue features in the predicted amino acid residue features, and the pathogenicity, pathogenic mechanism and drug sensitivity of unknown mutation sites under the same category are evaluated using the predicted amino acid residue features containing known mutation sites.

8. An epidermal growth factor receptor pathogenicity prediction device, characterized in that: include: Clinical mutation intersection analysis module, used to extract amino acid residue features based on the three-dimensional spatial conformation of compound mutation amino acid sequences; The point-surface conformation flexible fusion module is used to predict protein structural features based on the characteristics of amino acid residues; The point-surface conformation flexible fusion module includes a first prediction unit for predicting amino acid residue features using a convolutional layer of a first graph neural network; An amino acid cluster feature generating unit, used for generating amino acid cluster features according to predicted amino acid residue features; A second prediction unit, used to predict the amino acid cluster features using a convolutional layer of a second graph neural network; A protein subunit feature generation unit, used to generate protein subunit features based on predicted amino acid cluster features; A third graph convolution unit, for predicting protein subunit features using the convolutional layer of the third graph neural network; A protein structure feature generating unit, used for generating protein structure features according to protein subunit features; The point-surface conformation clustering dynamic evaluation module is used to evaluate the pathogenicity of mutation sites based on protein structural characteristics.

9. A server, characterized in that: The server comprises: one or more processors; a storage device for storing one or more programs, When the one or more programs are executed by the one or more processors, the one or more processors implement the method for predicting pathogenicity of epidermal growth factor receptor as described in any one of claims 1-7.

10. A storage medium comprising computer executable instructions, wherein the computer executable instructions are used to execute the method for predicting pathogenicity of epidermal growth factor receptor according to any one of claims 1 to 7 when executed by a computer processor.

Images