Individualized coding and non-coding cancer pathogenic gene identification method based on gene interaction network node centrality

Abstract

The invention discloses an individualized coding and non-coding cancer pathogenic gene identification method based on gene interaction network node centrality. The method comprises the following steps: constructing an individualized lncRNA-TF-mRNA gene interaction network for a patient through expression data of coding genes TF and mRNA of the patient and expression data of a non-coding gene lncRNA, and centrally evaluating a gene node on the established individualized lncRNA-TF-mRNA gene interaction network in combination with the gene interaction network of the gene node. Coding and non-coding cancer pathogenic genes of a patient are ranked to identify cancer pathogenic genes. The method has the beneficial effects that by implementing the method, the identification accuracy of individualized coding and non-coding cancer pathogenic genes is effectively improved.

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics and relates to an individualized coding and non-coding cancer pathogenic gene identification method based on the centrality of gene interaction network nodes. Background Art

[0002] Accurately identifying pathogenic genes is crucial to improving the accuracy of cancer diagnosis and optimizing treatment plans. Based on whether genes can produce protein-coding functions, they can be divided into two categories: coding genes and non-coding genes. Non-coding pathogenic genes have also been proven to be an important cause of cancer. However, existing personalized pathogenic gene identification methods mainly focus on finding coding pathogenic genes, and there are few methods to study the identification of non-coding cancer pathogenic genes. Using the coding and non-coding genomic data of cancer patients obtained by gene sequencing technology to identify the coding and non-coding cancer pathogenic genes of patients at the individual level has become an urgent need for personalized cancer diagnosis and treatment. Summary of the invention

[0003] The purpose of the present invention is to provide a method for identifying individualized coding and non-coding cancer pathogenic genes based on the centrality of gene interaction network nodes, so as to improve the accuracy of identifying individualized coding and non-coding cancer pathogenic genes.

[0004] The technical solution adopted by the present invention to achieve the above-mentioned purpose is:

[0005] The method for identifying individual coding and non-coding cancer pathogenic genes based on the centrality of gene interaction network nodes includes the following steps:

[0006] 1) Construction of personalized lncRNA-TF-mRNA gene interaction network:

[0007] For patient p and the coding gene expression data and non-coding gene lncRNA expression data of all patients in a given patient set, the patient's coding gene TF, mRNA and non-coding gene lncRNA are used as vertices, and the interactions between genes are used as edges to form the individualized lncRNA-TF-mRNA gene interaction network G = (V, E) of patient p, where V is the gene vertex set and E is the gene interaction edge set;

[0008] Assume A nⅹn is the adjacency matrix of the gene interaction network G, n is the number of gene vertices in G, and the absolute value of the Pearson correlation coefficient between the gene vertices of G is calculated based on the TF and mRNA expression data of the patient's coding genes and the lncRNA expression data of the non-coding genes as the gene interaction edge weight of G. When the edge weight between gene vertex i and gene vertex j is greater than or equal to the average value of all gene interaction edge weights in G, A ij =1, otherwise A ij=0;

[0009] Using the known protein-protein interaction, lncRNA-TF interaction and lncRNA-mRNA interaction information in the gene interaction database, filter out the edges between genes in the gene interaction network G that do not have interaction relationships to further optimize the patient's gene interaction network;

[0010] 2) Individualized cancer-causing gene sequencing and identification:

[0011] The gene network centrality of all genes in the personalized lncRNA-TF-mRNA gene interaction network G of patient p is iteratively updated using the following formula (1):

[0012] S m+1 =0.15*M T S m +0.85*S 0 (1)

[0013] Where M is the adjacency matrix A nⅹn The row normalized transfer matrix, S m and S m+1 represents the transition probability of each node in G at the mth step and m+1th step, S 0 represents the initial transition probability of each node in G, which is set to 1 / n; each element M of M is calculated using formula (2) ij ,i,j∈V:

[0014]

[0015] Where N(i) is all neighbor nodes of gene vertex i, DC(j) and DC(v) are the degrees of gene vertex j and gene vertex v; when the iteration of formula (2) reaches the maximum number of iterations 1000 times or the difference between the transition probabilities of the m+1th step and the mth step of formula (2) is less than 1e-6, the iteration of formula (1) stops;

[0016] Based on the gene network centrality obtained by formula (1), the final score of gene i is i It can be calculated by formula (3):

[0017]

[0018] β = 0.5 is the weight parameter, R (i) is the gene network centrality of gene i calculated by the gene interaction network G through formula (1), is the gene network connection score of gene i in the gene interaction network G, where θ jk is the number of shortest paths between gene j and gene k∈V, is the number of paths from gene j to gene k that pass through gene i; max{R(1),R(2),…,R ( n)} represents the maximum value of the gene network centrality of all genes in G calculated by formula (1), max{NC1,NC2,…,NC n} represents the maximum value of the gene network connection score of all genes in G;

[0019] Formula (3) is used to calculate the final scores of all genes in the gene interaction network G, and the coding and non-coding genes in G are ranked according to the final scores of the genes, and the genes with the highest ranking are output as candidate cancer pathogenic genes.

[0020] The present invention constructs an individualized lncRNA-TF-mRNA gene interaction network for patients based on the expression data of the patients' coding gene TF, mRNA and non-coding gene lncRNA, and on the established individualized lncRNA-TF-mRNA gene interaction network, the gene nodes are evaluated by combining the gene interaction network centrality of the gene nodes, and the coding and non-coding cancer pathogenic genes of the patients are sorted to identify cancer pathogenic genes.

[0021] The present invention has the beneficial effect that, compared with the existing methods, the method of the present invention effectively improves the accuracy of identifying individualized coding and non-coding cancer pathogenic genes. DETAILED DESCRIPTION

[0022] The method for identifying individualized coding and non-coding cancer pathogenic genes based on gene interaction network node centrality of the present invention comprises the following specific steps:

[0023] 1) Construction of personalized lncRNA-TF-mRNA gene interaction network:

[0024] For patient p and the coding gene expression data and non-coding gene lncRNA expression data of all patients in a given patient set, the patient's coding gene TF, mRNA and non-coding gene lncRNA are used as vertices, and the interactions between genes are used as edges to form the individualized lncRNA-TF-mRNA gene interaction network G = (V, E) of patient p, where V is the gene vertex set and E is the gene interaction edge set;

[0025] Assume A nⅹn is the adjacency matrix of the gene interaction network G, n is the number of gene vertices in G, and the absolute value of the Pearson correlation coefficient between the gene vertices of G is calculated based on the TF and mRNA expression data of the patient's coding genes and the lncRNA expression data of the non-coding genes as the gene interaction edge weight of G. When the edge weight between gene vertex i and gene vertex j is greater than or equal to the average value of all gene interaction edge weights in G, A ij=1, otherwise A ij =0;

[0026] Using the known protein-protein interaction, lncRNA-TF interaction and lncRNA-mRNA interaction information in the gene interaction database, filter out the edges between genes in the gene interaction network G that do not have interaction relationships to further optimize the patient's gene interaction network;

[0027] 2) Individualized cancer-causing gene sequencing and identification:

[0028] The gene network centrality of all genes in the personalized lncRNA-TF-mRNA gene interaction network G of patient p is iteratively updated using the following formula (1):

[0029] S m+1 =0.15*M T S m +0.85*S 0 (1)

[0030] Where M is the adjacency matrix A nⅹn The row normalized transfer matrix, S m and S m+1 represents the transition probability of each node in G at the mth step and m+1th step, S 0 represents the initial transition probability of each node in G, which is set to 1 / n; each element M of M is calculated using formula (2) ij ,i,j∈V:

[0031]

[0032] Where N(i) is all neighbor nodes of gene vertex i, DC(j) and DC(v) are the degrees of gene vertex j and gene vertex v; when the iteration of formula (2) reaches the maximum number of iterations 1000 times or the difference between the transition probabilities of the m+1th step and the mth step of formula (2) is less than 1e-6, the iteration of formula (1) stops;

[0033] Based on the gene network centrality obtained by formula (1), the final score of gene i is i It can be calculated by formula (3):

[0034]

[0035] β = 0.5 is the weight parameter, R (i) is the gene network centrality of gene i calculated by the gene interaction network G through formula (1), is the gene network connection score of gene i in the gene interaction network G, where θ jk is the number of shortest paths between gene j and gene k∈V, is the number of paths from gene j to gene k that pass through gene i; max{R (1) ,R (2) ,…,R (n)} represents the maximum value of the gene network centrality of all genes in G calculated by formula (1), max{NC1,NC2,…,NC n} represents the maximum value of the gene network connection score of all genes in G;

[0036] Formula (3) is used to calculate the final scores of all genes in the gene interaction network G, and the coding and non-coding genes in G are ranked according to the final scores of the genes, and the genes with the highest ranking are output as candidate cancer pathogenic genes.

[0037] Using the above method, the present invention was experimentally verified in five TCGA cancer data sets, namely: breast invasive cancer, colon adenocarcinoma, lung adenocarcinoma, prostate cancer and thyroid cancer. Tables 1 and 2 compare the recognition results of cancer pathogenic genes on the above five cancer data sets by the method proposed in the present invention and the existing cancer pathogenic gene identification method (see Yiran Huang, Fuhao Chen, Hongtao Sun, et al. Exploring gene-patient association to identify personalized cancer driver genes by linear neighborhood propagation [J]. BMC Bioinformatics, 2024, 25: 34). Table 3 is the number of known cancer pathogenic lncRNAs in the Lnc2Cancer v3.0 database identified by the method proposed in the present invention on five cancer data sets. The experimental results of Tables 1, 2 and 3 show that the method proposed in the present invention can accurately identify non-coding cancer pathogenic lncRNAs; compared with the existing methods, the method proposed in the present invention can more accurately identify coding and non-coding pathogenic genes of cancer.

[0038] Table 1 Comparison of the recognition accuracy of cancer pathogenic genes between the method of the present invention and the existing methods on five cancer data sets

[0039]

[0040] Table 2 Comparison of the recognition recall rate of cancer pathogenic genes between the method of the present invention and the existing methods on five cancer data sets

[0041]

[0042] Table 3 Number of known cancer pathogenic lncRNAs identified by the method of the present invention in the Lnc2Cancer v3.0 database

[0043]

Claims

1. A personalized method for identifying coding and non-coding cancer pathogenic genes based on the centrality of gene interaction network nodes, characterized in that: The steps include: 1) Construction of personalized lncRNA-TF-mRNA gene interaction network: For patient p and the coding gene expression data and non-coding gene lncRNA expression data of all patients in a given patient set, the patient's coding gene TF, mRNA and non-coding gene lncRNA are used as vertices, and the interactions between genes are used as edges to form the individualized lncRNA-TF-mRNA gene interaction network G = (V, E) of patient p, where V is the gene vertex set and E is the gene interaction edge set; Assume A nⅹn is the adjacency matrix of the gene interaction network G, n is the number of gene vertices in G, and the absolute value of the Pearson correlation coefficient between the gene vertices of G is calculated based on the TF and mRNA expression data of the patient's coding genes and the lncRNA expression data of the non-coding genes as the gene interaction edge weight of G. When the edge weight between gene vertex i and gene vertex j is greater than or equal to the average value of all gene interaction edge weights in G, A ij =1, otherwise A ij =0; Using the known protein-protein interaction, lncRNA-TF interaction and lncRNA-mRNA interaction information in the gene interaction database, filter out the edges between genes in the gene interaction network G that do not have interaction relationships to further optimize the patient's gene interaction network; 2) Individualized cancer-causing gene sequencing and identification: The gene network centrality of all genes in the personalized lncRNA-TF-mRNA gene interaction network G of patient p is iteratively updated using the following formula (1): S m+1 =0.15*M T S m +0.85*S 0 (1) Where M is the adjacency matrix A nⅹn The row normalized transfer matrix, S m and S m+1 represents the transition probability of each node in G at the mth step and m+1th step, S 0 represents the initial transition probability of each node in G, which is set to 1 / n; each element M of M is calculated using formula (2) ij ,i,j∈V: Where N(i) is all neighbor nodes of gene vertex i, DC(j) and DC(v) are the degrees of gene vertex j and gene vertex v; when the iteration of formula (2) reaches the maximum number of iterations 1000 times or the difference between the transition probabilities of the m+1th step and the mth step of formula (2) is less than 1e-6, the iteration of formula (1) stops; Based on the gene network centrality obtained by formula (1), the final score of gene i is i It can be calculated by formula (3): β = 0.5 is the weight parameter, R (i) is the gene network centrality of gene i calculated by the gene interaction network G through formula (1), is the gene network connection score of gene i in the gene interaction network G, where θ jk is the number of shortest paths between gene j and gene k∈V, is the number of paths from gene j to gene k that pass through gene i; max{R (1) ,R (2) ,…,R (n) } represents the maximum value of the gene network centrality of all genes in G calculated by formula (1), max{NC1,NC2,…,NC n } represents the maximum value of the gene network connection score of all genes in G; Formula (3) is used to calculate the final scores of all genes in the gene interaction network G, and the coding and non-coding genes in G are ranked according to the final scores of the genes, and the genes with the highest ranking are output as candidate cancer pathogenic genes.

Images