Cancer driver gene identification method and device based on multi-layer graph convolutional neural network

Abstract

The present invention discloses a method and device for identifying cancer driver genes based on a multi-layer graph convolutional neural network, including preprocessing and calculating the multi-omics features of the collected genes to obtain gene fusion features; using the collected genes as gene nodes and the functions between the genes as the edges between the gene nodes to obtain a gene network; obtaining network topology features based on the gene network; inputting the gene fusion features and the network topology features into two independent multi-layer graph convolutional neural networks for training; and sending the training results into two independent predictors to obtain the final prediction score of each gene. The present invention uses a multi-layer graph convolutional neural network to learn the features of the high-order neighbors of the nodes, thereby capturing information between genes of the same type; combining the initial residual and identity mapping techniques in each layer of the graph convolutional neural network; and using a random walk algorithm that supports depth-first and breadth-first searches to extract the global topology features of the network.

Description

Technical Field

[0001] The present invention relates to the technical field of cancer driver gene identification, and in particular to a cancer driver gene identification method and device based on a multi-layer graph convolutional neural network. Background Art

[0002] Cancer is caused by the accumulation of gene mutations. A small number of mutations provide selective growth advantages to mutated cells. Mutations that are critical to the development of cancer are considered driver mutations. Genes that carry driver mutations are called driver genes. Since the number of known cancer driver genes is limited, it is crucial to identify cancer driver genes. Correctly identifying cancer driver genes contributes to the progress of cancer diagnosis and cancer treatment. With the development of biotechnology and the implementation of projects such as The Canergenome Atlas (TCGA), International Cancer Genome Consortium (ICGC) and Catalogue Of Somatic Mutations In Cancer (COSMIC), a large amount of omics data is provided for the present invention, which helps the present invention design a computational method for identifying cancer driver genes. In this invention, the present application focuses on mutated driver genes.

[0003] Existing methods for identifying cancer driver genes can be roughly divided into three categories: methods based on gene mutation frequency, network-based methods, and machine learning-based methods. Methods based on gene mutation frequency, such as MutSig, MuSic, and MutSigCV, usually identify genes with mutation frequencies higher than the predefined background mutation rate as cancer driver genes. However, the inaccuracy of background mutation rate estimation and the neglect of low mutation frequency cancer driver genes will cause such methods to fail. Network-based methods assume that cancer driver genes work together in certain pathways and protein complexes. For example, HotNet2 and MUFFINN believe that cancer driver genes are often enriched in protein complexes or pathways. By projecting mutated genes onto protein-protein interaction (PPI) networks, they can identify highly significant mutation gene modules or essential genes that have a strong influence on other genes, thereby identifying driver genes. However, the reliability of PPI networks limits the identification effect of these network-based methods. In order to improve the reliability of PPI networks and make full use of valuable multi-omics data, researchers have developed a variety of methods to weight and filter PPI networks using multi-omics data, including methods to exclude noise connections based on constraints such as co-expression, co-subcellular localization, and co-tissue expression. However, these methods fail to effectively utilize the relationships between multi-omics data to improve the accuracy of driver gene prediction.

[0004] Recently, deep learning-based methods have made significant progress in the fields of molecular biology and genomics. In particular, several deep learning-based methods for identifying cancer driver genes have emerged, among which graph neural networks are particularly emphasized. Graph Convolutional Network is a novel network representation method that can learn node representations through network structure and node features. It is worth noting that most methods based on graph convolutional neural networks to identify cancer driver genes use shallow GCN architectures, which limits the model's ability to capture high-order neighbor features in the network. Therefore, a method using multi-layer GCNs is needed to learn high-order neighbor features of genes, thereby improving the accuracy of cancer driver gene identification. Summary of the invention

[0005] The present invention proposes a cancer driver gene identification method, device and storage medium based on a multi-layer graph convolutional neural network, which can solve at least one of the technical problems in the background technology.

[0006] To achieve the above object, the present invention adopts the following technical solutions:

[0007] A method for identifying cancer driver genes based on a multi-layer graph convolutional neural network, comprising the following steps:

[0008] Step 1: Preprocess and calculate the collected multi-omics features of genes to obtain gene fusion features;

[0009] Step 2: The collected genes are used as gene nodes, and the functions between genes are used as edges between gene nodes to obtain a gene network;

[0010] Step 3, obtaining network topology characteristics based on the gene network;

[0011] Step 4: Input the gene fusion features and network topology features into two independent multi-layer graph convolutional neural networks for training;

[0012] Step 5: Send the training results to two independent predictors to obtain the final prediction score of each gene.

[0013] Furthermore, the step 1 is specifically as follows:

[0014] First, the gene expression value of each gene was obtained by calculating the log2fold change between its expression value in cancer and the corresponding normal sample expression value, and then the average was taken for all samples;

[0015] Then, the gene mutation value of each gene was obtained by calculating the average of single nucleotide variations (SNVs) and copy number variations (CNAs) of all samples in the corresponding cancer type;

[0016] Secondly, the DNA methylation value is determined by the average value of the methylation signals of all cancer samples and normal samples. The specific calculation is calculated using formula (1):

[0017] (1)

[0018] in, Indicates the gene in the cth cancer i Differential DNA methylation values, and are the methylation signals in cancerous samples and paired normal samples, A sample set representing a type of cancer;

[0019] Finally, the calculated gene expression values, gene mutation values, DNA methylation values ​​and system-level features extracted by the tool are spliced ​​to obtain gene fusion features.

[0020] Furthermore, the step 2 specifically includes: collecting corresponding gene functions corresponding to corresponding genes in the database as edges.

[0021] Furthermore, the step 3 specifically includes: using the node2vec algorithm on the gene network to obtain the network topology characteristics of each gene;

[0022] Furthermore, the step 4 is specifically as follows: the two technologies of initial residual connection and identity mapping are added to the graph convolutional neural network model used, and the number of layers of the model is increased. The input is multi-omics features, network topology features and gene networks, and after passing through two multi-layer graph convolutional neural networks with the same structure, abstract multi-omics features and abstract network topology features are obtained;

[0023] Furthermore, the step 5 is specifically as follows: the abstract multi-omics features and the abstract network topology features are respectively sent to two independent predictors to obtain two prediction scores, and the two prediction scores are weightedly added to obtain the final prediction score of each gene.

[0024] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0025] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0026] It can be seen from the above technical solution that the cancer driver gene identification method and device based on multi-layer graph convolutional neural network of the present invention have the following beneficial effects:

[0027] 1. In the field of cancer driver gene identification, most research methods based on graph convolutional neural networks use shallow graph convolutional neural network architectures, which makes it difficult to directly learn the information between similar genes that are far apart in the network. To solve this problem, the present invention uses a multi-layer graph convolutional neural network to learn the characteristics of the high-order neighbors of the nodes, thereby capturing the information between similar genes.

[0028] 2. In the field of cancer driver gene identification, during the information aggregation process of graph convolutional neural networks, the unique features of driver genes may be smoothed by the features of their adjacent non-driver genes. In order to solve this problem, the present invention combines initial residual and identity mapping techniques in each layer of graph convolutional neural networks. On the one hand, these two technologies can alleviate the problem of feature over-smoothing caused by deepening the number of network layers; on the other hand, these two technologies can alleviate the problem that the unique features of driver genes may be smoothed by the features of their adjacent non-driver genes, because in the learning process of each layer of GCN, some original features will be added to ensure the uniqueness of the driver gene features.

[0029] 3. In the field of cancer driver gene identification, most current research methods use Deepwalk to extract network topological features. However, due to its algorithmic characteristics, Deepwalk is difficult to capture the global topological features of the network. Therefore, the present invention uses a random walk algorithm that supports depth-first and breadth-first search, namely Node2vec, to extract the global topological features of the network. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] Figure 1 It is a flowchart of the present invention;

[0031] Figure 2 AUROC value comparison diagram of this method and other seven methods in Example 1;

[0032] Figure 3 This is a comparison chart of the AUPR values ​​of this method and the other seven methods in Example 1;

[0033] Figure 4 AUROC value comparison diagram of this method and other four methods in Example 2;

[0034] Figure 5 This is a comparison chart of the AUPR values ​​of this method and the other four methods in Example 2. DETAILED DESCRIPTION

[0035] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments.

[0036] Example 1: The method described in the present invention is used to predict pan-cancer cancer driver genes, and the gene network comes from the KEGG and Reactome pathway databases (PathNet); the data of gene mutation values, DNA methylation values ​​and gene expression values ​​are from TCGA, involving 29,000 and 16 different cancer types (lung adenocarcinoma, lung squamous cell carcinoma, liver cancer, gastric cancer, colon cancer, esophageal cancer, rectal cancer, head and neck cancer, renal clear cell carcinoma, renal papillary cell carcinoma, thyroid cancer, endometrial cancer, cervical cancer, breast cancer, prostate cancer, bladder cancer).

[0037] The positive sample list was collected from NCG6.0 labeled with the cancer type. For the negative sample list, starting from all genes, genes in NCG, COSMIC, OMIM databases and KEGG cancer pathways were recursively removed. Finally, the resulting benchmark dataset included 796 positive samples and 2187 negative samples.

[0038] like Figure 1 As shown in the figure, the cancer driver gene identification method based on multi-layer graph convolutional neural network has the following specific steps:

[0039] Step 1: Perform relevant preprocessing and calculation on the multi-omics features of genes collected in the database and tools to obtain gene fusion features, such as Figure 1 As shown in (a);

[0040] First, the gene expression value of each gene was obtained by calculating the log2fold change between its expression value in cancer and the corresponding normal sample expression value, and then the average was taken for all samples;

[0041] Then, the gene mutation value of each gene was obtained by calculating the average of single nucleotide variations (SNVs) and copy number variations (CNAs) of all samples in the corresponding cancer type;

[0042] Secondly, the DNA methylation value is calculated by the average value of the methylation signals of all tumor samples and normal samples. The specific calculation is calculated using formula (1):

[0043] (1)

[0044] in, represents the differential DNA methylation value of gene i in the cth cancer, and are the methylation signals in cancerous samples and paired normal samples, A collection of samples representing a type of cancer.

[0045] Finally, the calculated gene expression values, gene mutation values, DNA methylation values ​​and system-level features extracted by the tool are spliced ​​to obtain the above-mentioned multi-omics features;

[0046] Step 2: The collected genes are used as gene nodes, and the functions between genes are used as edges between gene nodes to obtain a gene network. The gene network includes 7699 nodes and 92710 edges;

[0047] Step 3: Use the node2vec algorithm to obtain the network topology characteristics of each gene in the gene network, such as Figure 1 (b)

[0048] Step 4: Input the multi-omics features and network topology features into two independent multi-layer graph convolutional neural network models to perform feature learning on the multi-omics features and network topology features. Figure 1 As shown in (b);

[0049] Step 4.1. After adding the initial residual connection and identity mapping to the traditional graph convolutional neural network, the layered propagation rule of the graph convolutional neural network is shown in formula (2). Formula (2) is used to update the node feature representation to obtain the advanced multi-omics features. , denoted as Z B and advanced network topology features , denoted as Z T :

[0050] (2)

[0051] (3)

[0052] In formula (2), , , , represents the degree matrix, represents the unit matrix, represents the adjacency matrix; and are the hyperparameters of the initial residual connection and identity mapping, respectively, where , is the scale parameter set, Indicates the number of layers of the current graph neural network, represents a nonlinear activation function; Represents multi-omics features, Indicates that after Multi-omics features learned by layered graph convolutional neural networks; Indicates Multi-omics features learned by layered graph convolutional neural networks; It is Learnable weight matrices in layer graph convolutional neural networks;

[0053] In formula (3), Represents the network topology characteristics, Indicates that after Multi-omics features learned by layered graph convolutional neural networks; Indicates Multi-omics features learned by layered graph convolutional neural networks; It is The learnable weight matrix in the layer graph convolutional neural network, the other parameters are the same as formula (2);

[0054] Step 4.2: Use equations (4), (5) and (6) to construct the total loss function , and use The optimizer is used to update and calculate the parameters of the model in order to minimize the cross entropy loss function:

[0055] (4)

[0056] (5)

[0057] (6)

[0058] In formula (4) and formula (5), represents the true label of the gene, Represents the number of genes in the training set.

[0059] In formula (6), for The calculated cross entropy loss value is, for The calculated cross entropy loss value is, The scale parameter to be set.

[0060] Step 5: Calculate the prediction score , and the final prediction score ,like Figure 1 (c)

[0061] Step 5.1: Use formula (7) to obtain the prediction score of multi-omics features ;

[0062] (7)

[0063] In formula (7), is the activation function, is the dropout layer, is a fully connected layer;

[0064] Step 5.2: Use formula (8) to get the prediction score of network topology features ;

[0065] (8)

[0066] Step 5.3: Use formula (9) to convert the two prediction scores and The final prediction score is obtained after weighting :

[0067] (9)

[0068] In formula (8), is the advanced multi-omics feature Z B The calculated prediction score, It is the advanced network topology feature Z T The calculated prediction score, is the scale parameter set. The higher the output prediction score, the higher the probability that the gene is a cancer driver gene.

[0069] Example 2: The method described in the present invention is used to predict cancer driver genes for breast cancer, lung adenocarcinoma, bladder urothelial carcinoma and liver cancer. The protein interaction data comes from the consensus pathway database (CPDB); the gene mutation value, DNA methylation value and gene expression value data come from TCGA.

[0070] The positive sample list was collected from NCG6.0 labeled with the cancer type. For the negative sample list, starting from all genes, genes in NCG, COSMIC, OMIM databases and KEGG cancer pathways were recursively removed. Finally, the resulting benchmark dataset included 796 positive samples and 2187 negative samples.

[0071] like Figure 1 As shown in the figure, the cancer driver gene identification method based on multi-layer graph convolutional neural network has the following specific steps:

[0072] Step 1: Perform relevant preprocessing and calculation on the multi-omics features of genes collected in the database and tools to obtain gene fusion features, such as Figure 1 (a)

[0073] First, the gene expression value of each gene was obtained by calculating the log2fold change between its expression value in cancer and the corresponding normal sample expression value, and then the average was taken for all samples;

[0074] Then, the gene mutation value of each gene was obtained by calculating the average of single nucleotide variations (SNVs) and copy number variations (CNAs) of all samples in the corresponding cancer type;

[0075] Secondly, the DNA methylation value is calculated by the average value of the methylation signals of all tumor samples and normal samples. The specific calculation is calculated using formula (1):

[0076] (1)

[0077] in, represents the differential DNA methylation value of gene i in the cth cancer, and are the methylation signals in cancerous samples and paired normal samples, A collection of samples representing a type of cancer.

[0078] Finally, the calculated gene expression values, gene mutation values, DNA methylation values ​​and system-level features extracted by the tool are spliced ​​to obtain the above-mentioned multi-omics features;

[0079] Step 2: The collected genes are used as gene nodes, and the functions between genes are used as edges between gene nodes to obtain a gene network. The gene network includes 11,309 nodes and 621,988 edges;

[0080] Step 3: Use the node2vec algorithm to obtain the network topology characteristics of each gene in the gene network, such as Figure 1 (b)

[0081] Step 4: Input the multi-omics features and network topology features into two independent multi-layer graph convolutional neural network models to perform feature learning on the multi-omics features and network topology features. Figure 1 (b)

[0082] Step 4.1. After adding the initial residual connection and identity mapping to the traditional graph convolutional neural network, the layered propagation rule of the graph convolutional neural network is shown in formula (2). Formula (2) is used to update the node feature representation to obtain the advanced multi-omics features. , denoted as Z B and advanced network topology features , denoted as Z T :

[0083] (2)

[0084] (3)

[0085] In formula (2), , , , represents the degree matrix, represents the unit matrix, represents the adjacency matrix; and are the hyperparameters of the initial residual connection and identity mapping, respectively, where , is the scale parameter set, Indicates the number of layers of the current graph neural network, represents a nonlinear activation function; Represents multi-omics features, Indicates that after Multi-omics features learned by layered graph convolutional neural networks; Indicates Multi-omics features learned by layered graph convolutional neural networks; It is Learnable weight matrices in layer graph convolutional neural networks;

[0086] In formula (3), Represents the network topology characteristics, Indicates that after Multi-omics features learned by layered graph convolutional neural networks; Indicates Multi-omics features learned by layered graph convolutional neural networks; It is The learnable weight matrix in the layer graph convolutional neural network, the other parameters are the same as formula (2);

[0087] Step 4.2: Use equations (4), (5) and (6) to construct the total loss function , and use The optimizer is used to update and calculate the parameters of the model in order to minimize the cross entropy loss function:

[0088] (4)

[0089] (5)

[0090] (6)

[0091] In formula (4) and formula (5), represents the true label of the gene, Represents the number of genes in the training set.

[0092] In formula (6), for The calculated cross entropy loss value is, for The calculated cross entropy loss value is, The scale parameter to be set.

[0093] Step 5: Calculate the prediction score , and the final prediction score ,like Figure 1 (c)

[0094] Step 5.1: Use formula (7) to obtain the prediction score of multi-omics features ;

[0095] (7)

[0096] In formula (7), is the activation function, is the dropout layer, is a fully connected layer;

[0097] Step 5.2: Use formula (8) to get the prediction score of network topology features ;

[0098] (8)

[0099] Step 5.3: Use formula (9) to convert the two prediction scores and The final prediction score is obtained after weighting :

[0100] (9)

[0101] In formula (8), is the advanced multi-omics feature Z B The calculated prediction score, It is the advanced network topology feature Z T The calculated prediction score, is the scale parameter of the setting.

[0102] The higher the output prediction score, the higher the likelihood that the gene is a cancer driver gene.

[0103] In order to evaluate the performance of the multi-layer graph convolutional neural network (MLGCN), the proposed multi-layer graph convolutional neural network method was compared with the traditional machine learning and graph convolution methods proposed previously on three networks. AUROC and AUPR were selected as evaluation indicators, and comparative experiments were conducted after removing some of the strategies used. All experimental results were averaged after ten 5-fold cross validations.

[0104] AUROC values ​​for predicting pan-cancer cancer driver genes are as follows Figure 2 As shown, the AUPR value is Figure 3 shown.

[0105] from Figure 2 , Figure 3 It can be seen that the AUROC value and AUPR value of the MLGCN method are higher than those of the seven compared methods on the three networks, and the performance of MLGCN is very excellent. It can be seen that the MLGCN method of the present invention has a very obvious advantage in predicting cancer driver genes compared with other existing methods.

[0106] from Figure 4 , Figure 5 It can be seen that the AUROC value and AUPR value of the MLGCN method are higher than those of the four compared methods on the CPDB network, and the performance of MLGCN is very excellent. It can be seen that the MLGCN method of the present invention has a very obvious advantage over other existing methods in predicting single cancer driver genes.

[0107] On the other hand, the present invention further discloses a computer-readable storage medium storing a computer program, wherein when the computer program is executed by a processor, the processor executes the steps of the above method.

[0108] On the other hand, the present invention further discloses a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of the above method.

[0109] In another embodiment provided in the present application, a computer program product comprising instructions is also provided, which, when executed on a computer, enables the computer to execute any of the cancer driver gene identification methods based on a multi-layer graph convolutional neural network in the above embodiments.

[0110] It is understandable that the system, device and storage medium provided in the embodiments of the present invention correspond to the method provided in the embodiments of the present invention, and the explanation, examples and beneficial effects of the relevant contents can refer to the corresponding parts in the above methods.

[0111] In the above embodiments, it can be implemented in whole or in part by software, hardware, firmware or any combination thereof. When implemented using software, it can be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a server or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, server or data center. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or a data center that includes one or more available media integrated. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive Solid State Disk (SSD)), etc.

[0112] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the existence of other identical elements in the process, method, article or device including the elements.

[0113] Each embodiment in this specification is described in a related manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the system embodiment, since it is basically similar to the method embodiment, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiment.

[0114] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying cancer driver genes based on a multi-layer graph convolutional neural network, characterized in that: The following steps are included: Step 1: Preprocess and calculate the collected multi-omics features of genes to obtain gene fusion features; Step 2: The collected genes are used as gene nodes, and the functions between genes are used as edges between gene nodes to obtain a gene network; Step 3, obtaining network topology characteristics based on the gene network; Step 4: Input the gene fusion features and network topology features into two independent multi-layer graph convolutional neural networks for training; Step 5: Send the training results to two independent predictors to obtain the final prediction score of each gene; The step 1 is specifically as follows: First, the gene expression value of each gene was obtained by calculating the log2fold change between the expression value of the gene in cancer and the corresponding normal sample expression value, and then the average value was taken for all samples; Then, the gene mutation value of each gene was obtained by calculating the average of single nucleotide variations (SNVs) and copy number variations (CNAs) of all samples in the corresponding cancer type; Secondly, the DNA methylation value is determined by the average value of the methylation signals of all cancer samples and normal samples. The specific calculation is calculated using formula (1): (1) in, Indicates the gene in the cth cancer i Differential DNA methylation values, and are the methylation signals in cancerous samples and paired normal samples, A sample set representing a type of cancer; Finally, the calculated gene expression values, gene mutation values, DNA methylation values ​​and system-level features extracted by the tool are spliced ​​to obtain gene fusion features; Step 4 specifically includes: Step 4.1: After adding the initial residual connection and identity mapping to the traditional graph convolutional neural network, the layered propagation rule of the graph convolutional neural network is shown in formula (2). Formula (2) is used to update the node feature representation to obtain the advanced multi-omics features. , denoted as Z B and advanced network topology features , denoted as Z T : (2) (3) In formula (2), , , , represents the degree matrix, represents the unit matrix, represents the adjacency matrix; and are the hyperparameters of the initial residual connection and identity mapping, respectively, where , is the scale parameter set, Indicates the number of layers of the current graph neural network, represents a nonlinear activation function; Represents multi-omics features, Indicates that after Multi-omics features learned by layered graph convolutional neural networks; Indicates Multi-omics features learned by layered graph convolutional neural networks; It is Learnable weight matrices in layer graph convolutional neural networks; In formula (3), Represents the network topology characteristics, Indicates that after Multi-omics features learned by layered graph convolutional neural networks; Indicates Multi-omics features learned by layered graph convolutional neural networks; It is The learnable weight matrix in the layer graph convolutional neural network, the other parameters are the same as formula (2); Step 4.2: Use equations (4), (5) and (6) to construct the total loss function , and use The optimizer is used to update and calculate the parameters of the model in order to minimize the cross entropy loss function: (4) (5) (6) In formula (4) and formula (5), represents the true label of the gene, Represents the number of genes in the training set; In formula (6), for The calculated cross entropy loss value is, for The calculated cross entropy loss value is, The scale parameter to be set.

2. The method for identifying cancer driver genes based on a multi-layer graph convolutional neural network according to claim 1, characterized in that: In step 3, the node2vec algorithm is used on the gene network to obtain the network topological characteristics of each gene.

3. The method for identifying cancer driver genes based on a multi-layer graph convolutional neural network according to claim 1, characterized in that: Step 5 specifically includes calculating the prediction score , and the final prediction score ; Step 5.1: Use formula (7) to obtain the prediction score of multi-omics features ; (7) In formula (7), is the activation function, is the dropout layer, is a fully connected layer; Step 5.2: Use formula (8) to get the prediction score of network topology features ; (8) Step 5.3: Use formula (9) to convert the two prediction scores and The final prediction score is obtained after weighting : (9) In formula (8), is the advanced multi-omics feature Z B The calculated prediction score, It is the advanced network topology feature Z T The calculated prediction score, is the scale parameter of the setting.

4. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that: When the computer program is executed by the processor, the processor is caused to perform the steps of the method according to any one of claims 1 to 3.

Images