Cancer driver gene identification method based on biological functional network and pathway feature optimization

Abstract

The present invention relates to the field of gene identification technology, and in particular to a method for identifying cancer driver genes based on biological function network and pathway feature optimization, including: collecting and processing gene expression data and gene mutation data of several cancer patients, constructing a bipartite graph containing mutant genes and dysregulated expression genes; calculating the mutual exclusivity score of each pair of genes in the bipartite graph and weighting the bipartite graph to obtain a weighted bipartite graph; using a bidirectional random walk algorithm on the weighted bipartite graph to calculate the transfer probability score and coverage score after the bidirectional random walk; calculating the similarity score between patients based on the transfer probability score and the coverage score, and obtaining a list of driver genes for each patient based on the similarity score; calculating the cumulative score of each gene according to the list of driver genes, and identifying the genes ranked before the cumulative score as cancer driver genes. The present invention combines the heterogeneity and similarity in the molecular characteristics of patients for analysis to improve the accuracy of identifying driver genes.

Description

Technical Field

[0001] The present invention relates to the field of gene identification technology, and in particular to a method for identifying cancer driver genes based on biological function networks and pathway feature optimization. Background Art

[0002] Cancer, as a complex genetic variation disease with a high mortality rate, has plagued mankind for many years. It is well known that cancer is mainly caused by the gradual accumulation of various types of mutations. However, not all mutations play a key role in the occurrence and development of cancer. Mutations that are endowed with functional selection advantages and induce rapid growth of tumor cells in this process are called driver mutations, while those that have little or no effect on promoting cancer development are called passenger mutations. Genes that carry at least one driver mutation are called driver genes. Due to the heterogeneity and noise of mutations, it is challenging to identify the mutation signals of cancer driver genes from passenger genes that only carry passenger mutations, which is very important for understanding cancer biology and discovering therapeutic targets.

[0003] In recent years, with the advancement of cancer genome sequencing technology, projects such as The Cancer Genome Atlas (TCGA) and the International Cancer Genome Consortium (ICGC) have generated a large number of cancer-related omics datasets. The success of these projects has provided a rich resource for the development of cancer driver gene prediction methods. A classic method to distinguish driver genes from passenger genes involves calculating the background mutation rate of passenger genes. Then, genes that are statistically significantly higher than random mutations are considered driver genes. However, this method often ignores the fact that driver genes have low mutation frequencies and heterogeneity in many patients. In order to solve the problems caused by relying solely on the mutation frequency of individual genes, an effective method has emerged in recent years, which is to analyze the functional network or pathway formed by multiple interconnected genes. The network-based method is based on the assumption that the combination of candidate driver mutations may have an impact on a large number of differentially expressed genes in the functional interaction network. However, the network-based method is too dependent on known networks, which often have some incomplete and error-prone connections. The new pathway-based method overcomes the limitations of using prior knowledge by using somatic mutation patterns. They usually consider pathways with two characteristics of mutual exclusivity and high coverage in cancer to reduce the range of possible driver gene modules that need to be searched. High mutual exclusivity means that only one mutation is enough to disrupt a pathway, while high coverage means that a mutation in a pathway can be observed in a large number of patients. However, pathway-based methods consider all genes in a pathway to be equal, ignoring the topological interactions in large-scale signaling networks, while network-based methods can make up for this shortcoming.

[0004] In summary, both biological function networks and pathway algorithms obviously have their imperfections. However, they can complement each other to improve the accuracy of the identified candidate driver genes. In recent years, some methods have been developed to combine the mutual exclusivity and coverage information in interaction networks and pathways to identify cancer driver genes. However, these methods assume that the driver genes of all patients contribute equally to promoting cancer development, while ignoring the heterogeneity and commonality of gene mutations in patients. Therefore, a cancer driver gene identification method based on biological function network and pathway feature optimization is proposed. Summary of the invention

[0005] The purpose of the present invention is to provide a method for identifying cancer driver genes based on biological functional networks and pathway feature optimization, and to improve the accuracy and quality of identifying driver genes by combining heterogeneity and similarity in patient molecular characteristics for analysis.

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

[0007] Cancer driver gene identification methods based on biological function network and pathway feature optimization, including:

[0008] Collect gene expression data and gene mutation data from several cancer patients;

[0009] Processing the gene expression data and gene mutation data based on a biological function network to construct a bipartite graph including mutant genes and dysregulated expression genes;

[0010] Calculating the mutual exclusivity score of each pair of genes in the bipartite graph, and weighting the bipartite graph based on the mutual exclusivity score to obtain a weighted bipartite graph;

[0011] A bidirectional random walk algorithm is used on the weighted bipartite graph to calculate a transition probability score and a coverage score after the bidirectional random walk;

[0012] Calculating similarity scores between patients based on the transition probability scores and the coverage scores, and obtaining a driver gene list for each patient based on the similarity scores;

[0013] The cumulative score of each gene is calculated according to the driver gene list, and the top n genes ranked by the cumulative score are identified as cancer driver genes.

[0014] Optionally, processing the gene expression data and gene mutation data based on a biological function network to construct a bipartite graph containing mutant genes and dysregulated expression genes includes:

[0015] Processing the gene expression data into a dysregulated expression matrix by calculating the Z-score of gene expression;

[0016] The gene mutation data and the disordered expression matrix are processed into a bipartite graph based on a biological function network.

[0017] Optionally, processing the gene expression data into a dysregulated expression matrix by calculating the Z value of gene expression comprises:

[0018] If |Z-score|≥2, the gene is classified as the patient's dysregulated expression gene and assigned a value of 1 in the dysregulated expression matrix; otherwise, it is assigned a value of 0;

[0019] in:

[0020]

[0021] In the formula, exp l represents the expression value of gene l in all patients, μ l and δ l represent the mean and standard deviation of gene l in all patients, respectively.

[0022] Optionally, the mutual exclusivity score is 1-pvalue, where:

[0023]

[0024] Where pvalue represents the observed n under the condition of independent random permutation of gene 1 mutation and gene 2 mutation. 12 The probability of simultaneous mutation; n is the total number of pathways; n1 is the number of pathways carrying gene 1 mutation; n2 is the number of pathways carrying gene 2 mutation.

[0025] Optionally, the method for calculating the transition probability score is:

[0026]

[0027] In the formula, R m is the result of the first random walk; M a is the mutated gene vector of each patient; W ab is a weighted bipartite graph; R d is the result of the second random walk; D b is the vector of dysregulated expressed genes for each patient; α is the damping factor; Yes D b The transpose of YesR m The transpose of .

[0028] Optionally, the coverage score is obtained by the number of mutation gene enriched pathways of each patient after the bidirectional random walk.

[0029] Optionally, the method for calculating the similarity score between patients based on the transition probability score and the coverage score is:

[0030] SimPat ij =NS ij *CS ij

[0031] Where, SimPat ij represents the similarity score between patients i and j; NS ij represents the similarity score of the propagation of mutant genes to abnormally expressed genes and abnormally expressed genes to mutant genes between patients i and j; CS ij represents the coverage similarity score between patients i and j; represents the similarity score of the propagation of mutated genes to abnormally expressed genes between patients i and j; represents the similarity score of the propagation of abnormally expressed genes to mutated genes between patients i and j; Represents R m or R d The similarity of patients in dist(P i , P j ) is the Euclidean distance between patients i and j, P i and P j are patients i and j in R m or R d is a vector in; μ is a hyperparameter; β i，j Used to eliminate scaling problems; average(dist(P i , P i′ )) represents the average distance between patient i and each of its neighbors i′, P i′ is the neighbor i′ of patient i in R m or R d vector in; average(dist(P j , P j′ )) represents the average distance between patient j and each of its neighbors j′, P j′ is the neighbor j′ of patient j in R m or R d Vector in EP i represents the number of mutation gene enriched pathways of patient i; EP j represents the number of mutation gene-enriched pathways of patient j.

[0032] Optionally, the method for calculating the cumulative score of each gene according to the driver gene list is:

[0033]

[0034] Where H represents the number of patients, l represents the gene, SimPat ij represents the similarity score between patients i and j; represents the result of the first random walk of patient i; Indicates that patient i is in R m The average score results across all genes.

[0035] The beneficial effects of the present invention are:

[0036] The present invention takes into account the similarity and heterogeneity of each patient's molecular characteristics in the interaction network and pathway; in terms of the network, it measures how the mutant gene affects the gene expression of each patient in the weighted network; in the pathway, it measures the similarity between each pair of patients based on mutual exclusivity and coverage, complementing the heterogeneity and similarity in the patient's molecular characteristics, thereby improving the accuracy and quality of identifying driver genes. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for identifying cancer driver genes based on biological function network and pathway feature optimization according to an embodiment of the present invention;

[0039] Figure 2 This is a comparison chart of the recognition results Fscore of the method according to the embodiment of the present invention and DriverNet, Subdyquency, and EntroRank;

[0040] Figure 3 This is a comparison diagram of the pathway consistency between the method according to the embodiment of the present invention and DriverNet, Subdyquency, and EntroRank. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0043] This embodiment provides a method for identifying cancer driver genes based on biological function network and pathway feature optimization, including:

[0044] Collect gene expression data and gene mutation data from several cancer patients;

[0045] Gene expression data and gene mutation data are processed based on biological function networks to construct a bipartite graph containing mutant genes and dysregulated genes;

[0046] Calculate the mutual exclusivity score of each pair of genes in the bipartite graph, and weight the bipartite graph based on the mutual exclusivity score to obtain a weighted bipartite graph;

[0047] A bidirectional random walk algorithm is used on a weighted bipartite graph to calculate the transition probability score and coverage score after bidirectional random walk.

[0048] Calculate the similarity scores between patients based on the transition probability scores and the coverage scores, and obtain the driver gene list for each patient based on the similarity scores;

[0049] The cumulative score of each gene is calculated based on the driver gene list, and the top n genes ranked by cumulative scores are identified as cancer driver genes.

[0050] Furthermore, the gene expression data and gene mutation data of several cancer patients are collected as follows:

[0051] In this embodiment, RNA expression data and mutation data of four cancer types, lung cancer, breast cancer, head and neck squamous cell carcinoma (HNSC) and prostate cancer, were downloaded from the UCSC Xena website (https: / / xenabrowser.net / datapages / ). The website brings together genomic data from TCGA, GDC and other public cancer genome resources. The somatic mutation genes in the downloaded MAF file are integrated with non-synonymous mutations, ignoring mutations classified as "Silent", "5'UTR / 3'UTR" and other types. The corresponding cells in the mutation matrix are assigned a value of 1 to indicate that the somatic cells of a given sample have mutated. The gene transcription expression values ​​were preprocessed with log2(x+1) conversion in the form of RNA-seq HTseq count data, and an expression matrix was generated on the UCSC Xena website. Finally, only samples with both mutation data and expression data, i.e., patients, were retained.

[0052] The human functional interaction network (Fls) was used. The network has 14,034 nodes and 278,974 edges, including protein-protein interactions, gene co-expression, protein domain interactions, gene ontology (GO) annotations, and text-mined protein interactions.

[0053] After collecting patient data, cancer driver genes are identified for patients by integrating the propagation effects of patient-specific mutations in the biological functional network and the similarities between patients in terms of mutual exclusivity and coverage. Figure 1 As shown, the whole process mainly includes the following contents:

[0054] Step a, preprocessing the input expression data and mutation data in the biological function network, and constructing a bipartite graph containing mutant genes and dysregulated expression genes.

[0055] Step b, calculate the mutual exclusivity score of each pair of genes in the bipartite graph, weight the constructed bipartite graph, and construct a weighted bipartite graph.

[0056] In step c, a bidirectional random walk algorithm with restart is used on the weighted bipartite graph to quantify the impact of mutant genes and dysregulated genes, and the transition probability score and coverage score after the bidirectional random walk are obtained.

[0057] In step d, the similarity scores between patients are calculated by combining the transition probability scores after the bidirectional random walk and the coverage scores.

[0058] Step e, generating a patient prediction score list based on the combination of the transition probability score and the similarity score. The higher the gene ranking score of each patient, the greater the possibility that the gene will become the specific driver gene of the patient.

[0059] Furthermore, in step a, the input expression data and mutation data are preprocessed in the biological function network to construct a bipartite graph including mutant genes and dysregulated expression genes, which specifically includes:

[0060] Based on the view that the driver gene can affect the abnormal expression of downstream genes associated with it by regulating the biological function network, it is assumed that the driver gene can be identified by the abnormal expression events regulated in the biological function network. It is well known that the mutation of a single gene is sufficient to affect the abnormal changes of an entire pathway. Therefore, two interconnected genes rarely mutate at the same time in the same pathway. Based on this, this embodiment constructs a bipartite graph, and adjusts the weights of the edges in the network by calculating the mutual exclusivity scores between the associated nodes in the network.

[0061] Generate a bipartite graph consisting of mutant genes and dysregulated expression genes, and use formula (1) to calculate the Z value of gene expression:

[0062]

[0063] Among them, exp l represents the expression value of gene l in all patients, μ l and δ l Represent the mean and standard deviation of gene l in all patients, respectively. If |Z-score|≥2, the gene is classified as the dysregulated gene of the patient and assigned a value of 1 in the corresponding dysregulated expression matrix; otherwise, it is assigned a value of 0. Then a bipartite graph is generated, with one side of the bipartite graph being the mutant gene and the other side being the associated dysregulated expression gene. Specifically, for each patient, if there is a connection between the mutant gene l in the functional interaction network α and abnormally expressed genes b , then a link connecting the mutant gene l is constructed in the bipartite graph a and dysregulated gene expression b edge.

[0064] Furthermore, in step b, the mutual exclusivity score of each pair of genes in the bipartite graph is calculated to weight the constructed bipartite graph. The construction of the weighted bipartite graph specifically includes:

[0065] Assign the edge weights of the mutual exclusivity score to the previously constructed bipartite graph. Based on the principle of mutual exclusivity, it is assumed that within the same pathway, a mutant gene and its related dysregulated gene are unlikely to mutate at the same time. Therefore, the calculation of the mutual exclusivity score is shown in function (2):

[0066]

[0067] Among them, pvalue represents the observed n under independent random permutations of gene 1 mutation and gene 2 mutation. 12 The probability of simultaneous mutation; n is the total number of pathways; n1 is the number of pathways carrying mutations in gene 1; n2 is the number of pathways carrying mutations in gene 2. This function relies on Fisher's exact test and uses random permutations of mutant gene pairs as the null model to assess whether the frequency of double mutations is significantly lower than expected. The smaller the pvalue, the greater the possibility that the two genes have common mutations in the common pathway. Since the mutual exclusivity score is the opposite of the pvalue, it is expressed as 1 minus the pvalue. Subsequently, the mutual exclusivity score is normalized using the Sigmoid function (3) based on the node degree in the network to reduce the impact of the node degree on the final result.

[0068]

[0069] Where D(·) represents the mutant gene l a and dysregulated gene expression b degree; λ∈(0,1) is an adjustable parameter.

[0070] Furthermore, in step c, a bidirectional random walk algorithm with restart is used on the weighted bipartite graph to quantify the impact of mutant genes and dysregulated genes, and the transition probability score and coverage score after the bidirectional random walk are obtained, which specifically include:

[0071] The restarted random walk algorithm is a ranking algorithm widely used in various scenarios such as bipartite graphs. During the random walk process, the walker starts from a node of one type (e.g., type A) and transfers to another type of connected node (e.g., type B) in a probabilistic manner, or randomly jumps to any node in type A. The algorithm comprehensively considers the patterns and relationships between nodes of different types and effectively reveals the importance or role of each node in the bipartite graph. Therefore, the bidirectional random walk algorithm with restart is used to learn the effects of somatic mutations and dysregulated expression genes in the weighted bipartite graph from the bidirectional diffusion process. First, the diffusion effect from the mutant gene to its connected dysregulated expression genes is calculated using equation (4). For each patient, if the corresponding gene ι is mutated, the initial value of the mutant gene is assigned to 1. Similarly, if the corresponding gene ι is not mutated while showing abnormal expression, the initial value of the dysregulated expression gene is set to 2. Then, the result of the first random walk is used as the initial value of the mutant gene in the next round. At the same time, the diffusion effect from the abnormally expressed gene to its connected mutant gene, that is, the transition probability score, is calculated using equation (5):

[0072]

[0073] Among them, M α is the mutated gene vector of each patient; W ab is a weighted bipartite graph; D b is the vector of dysregulated expressed genes for each patient; α is the damping factor.

[0074] The coverage score was calculated by the number of pathways enriched for mutation genes in each patient after the bidirectional random walk.

[0075] Furthermore, in step d, calculating the similarity score between patients by combining the coverage score and the transition probability score includes:

[0076] First, for each pair of patients, it is assumed that if the mutant gene of one patient has a larger number of similar dysregulated genes connected to the mutant gene of the other patient, and vice versa, if the dysregulated gene of one patient has a larger number of similar mutant genes connected to the dysregulated gene of the other patient, then the two patients are more likely to have similar driver genes. The similarity of the patients is calculated from the mutually exclusive weighted bipartite graph after the bidirectional random walk using equation (6). The similarity of the propagation of mutant genes to abnormally expressed genes and the propagation of abnormally expressed genes to mutant genes is calculated separately.

[0077]

[0078] in, Represents R m or P d The similarity of patients in dist(P i , P j ) is the Euclidean distance between patients i and j; μ is a hyperparameter ranging from (0.3, 0.8); β ij It is used to eliminate the scaling problem and is defined as shown in equation (7).

[0079]

[0080] Among them, average(dist(P i , P i′ )) represents the average distance between patient i and each of its neighbors i′; average(dist(P j , P j′ )) represents the average distance between patient j and each of its neighbors j′. Subsequently, the patient similarity in the entire network was normalized using equation (8).

[0081]

[0082] Subsequently, the normalized patient similarities between the propagation of mutant genes to aberrantly expressed genes and the propagation of aberrantly expressed genes to mutant genes were summed using equation (9).

[0083]

[0084] In the second step, based on the principle that mutations within a pathway may co-occur in a large number of patients, the similarity of the patients in the pathway coverage score was calculated. The Jaccard index was used to measure the similarity of the pathways affected by the mutated gene in each patient, see equation (10).

[0085]

[0086] Among them, CS ijrepresents the coverage similarity score between patients i and j; EP i represents the number of mutation gene enriched pathways of patient i; EP j represents the number of mutation gene-enriched pathways of patient j.

[0087] All similarity scores are then combined using the following equation (11):

[0088] SimPat ij =NS ij *CS ij (11)

[0089] Furthermore, in step e, generating a patient prediction score list according to a combination of the transition probability score and the similarity score includes:

[0090] The ranking value of the driver gene is calculated using a combination of the patient similarity score and the transition probability score after random walk of the mutant gene, see Equation (12).

[0091]

[0092] Where H represents the number of patients, l represents the gene, SimPat ij represents the similarity score between patients i and j; represents the result of the first random walk of patient i, spreading from the mutant gene to the connected abnormally expressed genes; Indicates that patient i is in R m The average score results across all genes.

[0093] A list of driver genes was obtained for each patient. Then, the cumulative score of each gene in all patients was calculated based on the comprehensive score of equation (12). Genes with higher cumulative scores are more likely to be identified as driver genes, indicating that they play a role in promoting the development of cancer.

[0094] The method proposed in this embodiment is evaluated as follows:

[0095] (1) Evaluation indicators;

[0096] To evaluate the performance of this method, a comprehensive evaluation framework was constructed, including prediction of known cancer genes (i.e., Fscore), literature mining analysis, and functional enrichment analysis. The Network of Cancer Genes (NCGs) version 7.0 was used as a reference for known cancer driver genes, which includes 3,366 cancer driver genes that have been well confirmed or predicted in 122 cancer types.

[0097] First, we calculated Fscore, which is the harmonic mean of precision and recall, to measure the consistency between the driver genes predicted by this method and the known driver genes in NCGs 7.0. This metric is crucial for evaluating the ability of the model to identify true cancer driver genes, see formulas (13)-(15).

[0098]

[0099] In addition, literature mining and functional enrichment analysis were performed, using the DAVID online database for gene ontology (GO) and Kyoto Encyclopedia of Genomes (KEGG) pathway enrichment analysis. DAVID is a web-based analysis tool that integrates biological functions extracted from large lists of genes and proteins. In the GO enrichment analysis, three enriched gene ontology sets were focused on: COTERM_BO_DIRECT, GOTERM_CC_DIRECT, and GOTERM_MF_DIRECT. The significance of enrichment was set at a p-value cutoff of 0.05. In the literature mining analysis, the PubTator database was used, which is a tool that can identify the co-occurrence of predicted driver genes and cancer-related keywords in the literature.

[0100] (2) comparative evaluation;

[0101] Based on the RNA expression and mutation data of four cancer types collected from TCGA, including lung cancer, breast cancer, head and neck squamous cell carcinoma (HNSC), and prostate cancer, this method (MCSdriver) was compared and evaluated with DriverNet, Subdyquency, and EntroRank cancer driver gene identification methods using the above evaluation indicators.

[0102] A. Fscore;

[0103] The recognition results of MCSdriver were compared with those of DriverNet, Subdyquency, EntroRank and other methods, and the Fscore of the top 200 driver genes in each cancer type was calculated, such as Figure 2 Clearly, our approach consistently shows the best performance across all cancer types.

[0104] B. Pathway and gene ontology enrichment analysis;

[0105] The top N (N∈{50,100,150,200}) driver genes identified by MCSdriver and DriverNet, Subdyquency, and EntroRank were taken and their consistency with the reference driver genes in the significantly enriched pathways was evaluated. This consistency was quantified by the ratio of the intersection of the enriched pathway terms of the ranked gene set and the reference gene set to the total number of enriched pathways in the ranked gene set. Figure 3 The pathway consistency values ​​of MCSdriver, DriverNet, Subdyquency, and EntroRank in four cancer types are shown. Notably, MCSdriver showed high consistency values, with average scores above 0.7 in most cancer types, indicating a high degree of overlap with the reference driver gene set. MCSdriver performed well on all total values ​​in prostate cancer compared to other methods.

[0106] The embodiments described above are only descriptions of the preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Without departing from the design spirit of the present invention, various modifications and improvements made to the technical solutions of the present invention by ordinary technicians in this field should all fall within the protection scope determined by the claims of the present invention.

Claims

1. A method for identifying cancer driver genes based on biological function networks and pathway feature optimization, characterized in that: include: Collect gene expression data and gene mutation data from several cancer patients; Processing the gene expression data and gene mutation data based on a biological function network to construct a bipartite graph including mutant genes and dysregulated expression genes; Calculating the mutual exclusivity score of each pair of genes in the bipartite graph, and weighting the bipartite graph based on the mutual exclusivity score to obtain a weighted bipartite graph; A bidirectional random walk algorithm is used on the weighted bipartite graph to calculate a transition probability score and a coverage score after the bidirectional random walk; Calculating similarity scores between patients based on the transition probability scores and the coverage scores, and obtaining a driver gene list for each patient based on the similarity scores; Calculating the cumulative score of each gene according to the driver gene list, and identifying the top n genes ranked by the cumulative score as cancer driver genes; The method for calculating the transition probability score is: In the formula, R m is the result of the first random walk; M a is the mutated gene vector of each patient; W ab is a weighted bipartite graph; R d is the result of the second random walk; D b is the vector of dysregulated expressed genes for each patient; α is the damping factor; Yes D b The transpose of YesR m The transpose of The method for calculating the similarity score between patients based on the transition probability score and the coverage score is: SimPat ij =NS ij *CS ij Where, SimPat ij represents the similarity score between patients i and j; NS ij represents the similarity score of the propagation of mutant genes to abnormally expressed genes and abnormally expressed genes to mutant genes between patients i and j; CS ij represents the coverage similarity score between patients i and j; represents the similarity score of the propagation of mutated genes to abnormally expressed genes between patients i and j; represents the similarity score of the propagation of abnormally expressed genes to mutated genes between patients i and j; Represents R m or R d The similarity of patients in dist(P i , P j ) is the Euclidean distance between patients i and j, P i and P j are patients i and j in R m or R d is a vector in; μ is a hyperparameter; β i，j Used to eliminate scaling problems; average(dist(P i , P i' )) represents the average distance between patient i and each of its neighbors i', P i′ is the neighbor i′ of patient i in R m or R d vector in; average(dist(P j , P j′ )) represents the average distance between patient j and each of its neighbors j′, P j′ is the neighbor j′ of patient j in R m or R d Vector in EP i represents the number of mutation gene-enriched pathways of patient i; EP j represents the number of mutation gene-enriched pathways of patient j; The method for calculating the cumulative score of each gene based on the driver gene list is: Where H represents the number of patients, l represents the gene, SimPat ij represents the similarity score between patients i and j; represents the result of the first random walk of patient i; Indicates that patient i is in R m The average score results across all genes.

2. The method for identifying cancer driver genes based on biological function network and pathway feature optimization according to claim 1, characterized in that: Processing the gene expression data and gene mutation data based on the biological function network to construct a bipartite graph containing mutant genes and dysregulated expression genes includes: Processing the gene expression data into a dysregulated expression matrix by calculating the Z-score of gene expression; The gene mutation data and the disordered expression matrix are processed into a bipartite graph based on a biological function network.

3. The method for identifying cancer driver genes based on biological function network and pathway feature optimization according to claim 2, characterized in that: Processing the gene expression data into a dysregulated expression matrix by calculating the Z-score of gene expression includes: If |Z-score|≥2, the gene is classified as the patient's dysregulated expression gene and assigned a value of 1 in the dysregulated expression matrix; otherwise, it is assigned a value of 0; in: In the formula, exp l represents the expression value of gene l in all patients, μ l and δ l represent the mean and standard deviation of gene l in all patients, respectively.

4. The method for identifying cancer driver genes based on biological function network and pathway feature optimization according to claim 1, characterized in that: The mutual exclusivity score is 1-pvalue, where: Where pvalue represents the observed n under the condition of independent random permutation of gene 1 mutation and gene 2 mutation. 12 The probability of simultaneous mutation; n is the total number of pathways; n1 is the number of pathways carrying gene 1 mutation; n2 is the number of pathways carrying gene 2 mutation.

5. The method for identifying cancer driver genes based on biological function network and pathway feature optimization according to claim 1, characterized in that: The coverage score is obtained by the number of mutation gene enriched pathways for each patient after the bidirectional random walk.

Images