A multi-omics cancer sample representation method based on graph representation and related device

Abstract

The application discloses a kind of multi-omics cancer sample representation methods based on graph representation and related devices, representation method includes: obtaining gene public library data;Gene public library data is handled, and the gene expression data and methylation data of the variance of cancer sample is greater than threshold value are obtained;After the gene expression data and methylation data of the variance of cancer sample is greater than threshold value are handled, the multi-omics data of cancer sample is obtained, the intersection of the multi-omics data of cancer sample is obtained, and the common gene is obtained;Based on common gene, the gene interaction network of corresponding gene is obtained through protein interaction network database, after filtering out part of scatter point in gene interaction network and integration, the sample gene interaction network skeleton is obtained;The multi-omics data of cancer sample and sample gene interaction network skeleton representation are integrated using sample graph representation algorithm.The application processes multi-omics data, obtains the omics data corresponding to the gene highly related to cancer occurrence, so as to reduce the influence of experimental bias in measurement and sequencing data on subsequent conclusion.

Description

Technical Field

[0001] The present invention belongs to the technical field of bioinformatics, and in particular relates to a multi-omics cancer sample representation method based on graph representation and related devices. Background Art

[0002] Breast cancer, lung cancer, and kidney cancer are common and clinically harmful malignant tumors that pose a serious threat to global health. According to a study on the domestic cancer situation published in JNCC in 2022, breast cancer is one of the most common cancers in women, and lung cancer is one of the cancers that causes the most deaths worldwide. The incidence and mortality of kidney cancer are also increasing year by year. The reason is that cancer is a malignant tumor with high genetic heterogeneity. These tumor cells carrying different gene mutation spectra and genetic phenotypes have different biological behaviors and sensitivities to treatment. In addition, the same type of tumor may have genetic variations between different patients, resulting in great differences in the response to therapeutic drugs.

[0003] With the rapid advancement of high-throughput biotechnology and artificial intelligence, researchers are increasingly interested in integrating and analyzing multi-omics data from multiple cancer datasets. Machine learning methods are widely used in cancer research, providing many new tools to integrate and analyze diverse omics data to obtain more accurate predictions and advance precision medicine.

[0004] Among the current mainstream multi-omics data fusion methods, there are three main approaches: feature-linked multi-omics data integration, model-based multi-omics data integration, and network-based multi-omics methods. The first is the feature-linked multi-omics data integration method. This method directly combines omics data, including proteomics, genomics, and transcriptomics, to form a sample union matrix containing multi-omics data. This union matrix is ​​then explored using matrix factorization, supervised, and unsupervised methods. As research deepens, researchers have gradually recognized this limitation and have begun to seek more flexible and effective methods. Consequently, model-based multi-omics data integration methods have been proposed. This method integrates independent models constructed from different single-omics models into a final model. Ensemble learning methods have shown significant advantages in improving model generalization, stability, and flexibility. Finally, network-based multi-omics methods combine multi-omics data into a network to model biological laws. This method accurately reflects the interactions between omics data and also reveals the biological processes that influence each other between nodes within individual omics.

[0005] Among the most active approaches, network-based multi-omics approaches employ two main approaches. The first approach uses genes as nodes and gene interactions as edges, utilizing various omics data from genes as node features. Graph neural networks are then used to perform node-level predictions, such as predicting cancer genes, or edge-level tasks, such as predicting interactions between potential unknown genes. This approach, which uses genes as nodes and gene interactions as edges, produces a meaningless overall graph, enabling only node-level predictions and failing to infer individual samples. The second approach focuses on samples, using cancer samples as nodes. High-dimensional omics data from these samples is compressed and concatenated into a matrix, serving as node features. Similarities between samples form the edges between nodes. Subsequently, graph neural networks are used to perform node-level predictions, such as predicting sample survival or genotype. This approach, which uses samples as nodes and similarities between samples as edges, does not effectively model the network topology of multiple genes. Current multi-omics fusion approaches based on network models have limitations.

[0006] It can be seen that the existing sample characterization methods are still unable to construct a special data structure that can not only model the topological relationship of multi-gene networks but also perform downstream predictions for individual samples. Summary of the Invention

[0007] In order to overcome the technical problem in the existing technology that the network topological relationship between multiple genes cannot be understood, and thus downstream predictions cannot be made, the present invention aims to propose a multi-omics cancer sample representation method and related devices based on graph representation. This sample representation method can comprehensively understand the multi-gene network topological relationship and realize fine-grained representation of single samples. Therefore, using this sample representation method, more accurate graph classification can be obtained, and functions such as molecular subtype prediction and drug effect prediction of cancer samples can be realized.

[0008] To achieve the above object, the technical solution adopted by the present invention is as follows:

[0009] A first aspect of the present invention provides a method for representing multi-omics cancer samples based on graph representation, comprising the following steps:

[0010] Access to public gene bank data;

[0011] Processing the public gene library data to obtain gene expression data and methylation data of cancer samples with variance greater than the threshold;

[0012] After processing the gene expression data and methylation data of cancer samples with variance greater than a threshold, multi-omics data of cancer samples are obtained, and the intersection of the multi-omics data of cancer samples is taken to obtain common genes;

[0013] Based on public genes, the gene interaction network of the corresponding gene is obtained through the protein interaction network database. After filtering out some scattered points in the gene interaction network and integrating them, the sample gene interaction network skeleton is obtained;

[0014] The multi-omics data of cancer samples and the sample gene interaction network skeleton representation are integrated using a sample graph representation algorithm to obtain a multi-omics cancer sample representation based on graph representation.

[0015] Furthermore, the methylation level of the methylation data was calculated using the following formula:

[0016]

[0017] Where, represents the methylation level of methylation data, Indicates the methylation level of the gene body region, Gene body represents the gene body region, i represents the gene number, and Median represents the median.

[0018] Furthermore, the public genes are obtained by the following formula:

[0019] G=G Rk(Var(pro_meth))＜301 ∩G Rk(Var(body_meth))＜301 ∩G Rk(Var(Exp))＜301 (3)

[0020] Where G is a common gene, Rk is the variance ranking of genomic data, Var is the variance, pro_meth is the methylation level of the promoter region, body_meth is the methylation level of the gene body region, Exp is the gene expression level, G Rk(Var(pro_meth))＜301 The top 300 genes with the largest variance in their promoter regions are G Rk(Var(body_meth))＜301 G is the top 300 genes with the largest variance in the genome region. Rk(Var(Exp))＜301 The top 300 genes in terms of gene expression level variance.

[0021] Furthermore, the gene expression data and methylation data of cancer samples with variance greater than a threshold were processed using R scripts.

[0022] Furthermore, edges with gene interaction confidence greater than 0.4 were selected as the skeleton of the sample gene interaction network.

[0023] Furthermore, the multi-omics cancer sample representation S based on graph representation is:

[0024] S={(G1,y1),(G2,y2),(G3,y3)...(G N ,y N )}

[0025] Among them, S represents a certain cancer type graph data set, Gi represents the i-th sample in the dataset, y i Represents the cancer subtype label corresponding to the i-th sample.

[0026] Furthermore, the multi-omics data of cancer samples and the sample gene interaction network skeleton representation are integrated using a sample graph representation algorithm to obtain a graph-based multi-omics cancer sample representation, including the following steps:

[0027] The sample gene interaction network skeleton was reconstructed and represented as a two-dimensional list, where rows represent source nodes, columns represent target nodes, and the values ​​in the list represent whether there is an edge. Using the sample gene interaction network skeleton, it is possible to determine whether there is an edge between two genes; if there is an edge, the adjacent node number of the corresponding node is used as the feature of this node; otherwise, the search will continue until the entire two-dimensional list is traversed;

[0028] Then, for each node, the corresponding three genomic data are integrated, and the integration results are used as the features of this node to obtain a multi-omics cancer sample representation based on graph representation.

[0029] A second aspect of the present invention provides a multi-omics cancer sample representation system based on graph representation, comprising:

[0030] Data acquisition module, used to obtain gene public library data;

[0031] A data processing module is used to process the gene public library data to obtain gene expression data and methylation data of cancer samples with variance greater than a threshold;

[0032] The public gene acquisition module is used to process the gene expression data and methylation data of cancer samples with variance greater than a threshold to obtain multi-omics data of cancer samples. The multi-omics data of cancer samples are intersected to obtain public genes.

[0033] The sample gene interaction network skeleton acquisition module is used to obtain the gene interaction network of the corresponding gene based on the public gene through the protein interaction network database, filter out some scattered points in the gene interaction network and integrate them to obtain the sample gene interaction network skeleton;

[0034] The integration module is used to integrate the multi-omics data of cancer samples with the sample gene interaction network skeleton representation using a sample graph representation algorithm to obtain a multi-omics cancer sample representation based on graph representation.

[0035] A third aspect of the present invention provides an electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the multi-omics cancer sample representation method based on graph representation.

[0036] A fourth aspect of the present invention provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements a multi-omics cancer sample representation method based on graph representation.

[0037] Compared with the prior art, the present invention has the following technical effects:

[0038] First, in the present invention, by processing the gene public library data, gene expression data and methylation data of cancer samples with variance greater than a threshold are obtained; after processing the gene expression data and methylation data of cancer samples with variance greater than a threshold, multi-omics data of cancer samples are obtained, and the multi-omics data of cancer samples are intersected to obtain common genes. The new high-dimensional multi-omics data processing scheme for cancer samples can be used to process the multi-omics data to obtain omics data corresponding to genes that are highly correlated with cancer occurrence, thereby reducing the impact of experimental bias in measurement and sequencing data on subsequent conclusions. Based on the new high-dimensional multi-omics data processing scheme for cancer samples, the present invention represents the high-dimensional multi-omics data of cancer as the features of nodes in the graph, effectively achieving dimensionality reduction of multi-omics data. Secondly, verification was carried out based on this sample representation method, proving that the cancer sample representation method proposed by the present invention can effectively perform downstream analysis and become an effective tool for scientific researchers to analyze cancer sample data. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] Figure 1 Provide an overall flowchart for cancer omics data processing;

[0040] Figure 2 Constructing module (I) for the sample graph model;

[0041] Figure 3 Building modules for the sample graph model (II);

[0042] Figure 4 This is an example of the application of graph-based representation of cancer omics data;

[0043] Figure 5 Figure 1 shows the prediction results of molecular subtypes of breast cancer using this framework.

[0044] Figure 6 Schematic diagram of the graph-based multi-omics cancer sample representation system. DETAILED DESCRIPTION

[0045] To facilitate understanding of the present invention, the present invention will be described more fully below with reference to the accompanying drawings. The accompanying drawings illustrate preferred embodiments of the present invention. However, the present invention may be implemented in a variety of different forms and is not limited to the embodiments described herein. Rather, these embodiments are provided to provide a more thorough and comprehensive understanding of the disclosure of the present invention.

[0046] See also Figure 1 The multi-omics cancer sample representation method based on graph representation of the present invention comprises the following steps:

[0047] 1) Process high-variance gene expression data and methylation data of cancer samples obtained from public databases (here using the UCSC Xena database as an example). After processing with R scripts, three types of genomic data (i.e., sample expression matrix, promoter region methylation matrix, and gene body region methylation matrix) are obtained, i.e., multi-omics data of cancer samples.

[0048] Current research indicates that methylation in different DNA regions influences gene expression differently. The mechanisms of gene expression influenced by methylation at different sites within the DNA molecule are complex. Studies have shown that DNA methylation at promoters and promoter regions inhibits gene expression, while DNA methylation within the gene body may promote gene expression. Further research has revealed that the relationship between DNA methylation within the gene body and gene expression is correlated with its location within CpG islands (CGIs). In the normal human genome, CpGs within CpG islands are typically unmethylated, while CpGs outside CpG islands are typically methylated. Furthermore, research has shown that DNA methylation is closely linked to the development of cancer. Following tumorigenesis, CpG islands typically become hypermethylated, while the genome exhibits hypomethylation. Hypermethylation leads to decreased expression of tumor suppressor genes, thereby enhancing cancer cell formation and survival. Hypomethylation, on the other hand, can lead to decreased methylation levels in oncogenes, leading to their activation, as well as activation of some transposons, resulting in decreased chromosomal stability and ultimately contributing to the development of malignant tumors. Therefore, for the methylated regions that inhibit gene expression, represented by the promoter region, the methylation status of the gene in these regions is obtained according to formula (1):

[0049]

[0050] in, Indicates the methylation level of the DNA methylation region that inhibits the expression of gene i, that is, the methylation level of the promoter region. Indicates the methylation status of the upstream region from the transcription start site of the gene. Indicates the methylation status of the first exon region of the gene. Indicates the methylation status of the 5' untranslated region of the gene. The methylation status of the upstream region from the transcription start site of the gene. The reason for taking the median is that methylation data often presents an asymmetric distribution or contains outliers. Therefore, taking the median can better represent the central trend of the data and reflect the true methylation level of the sample.

[0051] For the methylated regions that promote gene expression, represented by the gene body region, the present invention processes them according to formula (2), wherein Indicates the methylation level of the DNA methylation region that promotes the expression of gene i, that is, the methylation level of the gene body region body_meth.

[0052]

[0053] Where, represents the methylation level of methylation data, Indicates the methylation level of the gene body region, Gene body represents the gene body region, i represents the gene number, and Median represents the median.

[0054] See also Figure 2 The molecular omics data of samples obtained from the public cancer sample database TCGA (The Cancer Genome Atlas) were used as the data source. The omics data pre-processed in the UCSC Xena database were used to extract the gene expression data files and methylation data files HM450 files of the samples. The gene expression data files of the samples were subjected to variance filtering to obtain the sample expression matrix. The methylation data files HM450 files and the CHAMP probe-gene name package in the R language were subjected to variance filtering to obtain the promoter region methylation matrix and the gene body region methylation matrix. The columns of the three matrices, the sample expression matrix, the promoter region methylation matrix and the gene body region methylation matrix, represent one gene, and the rows represent one sample. Then, according to formula (3), the top 300 genes with the largest variance in the three matrices were selected.

[0055] G=G Rk(Var(pro_meth))＜301 ∩G Rk(Var(body_meth))＜301 ∩G Rk(Var(Exp))＜301 (3)

[0056] Where G is a common gene, Rk is the variance ranking of genomic data, Var is the variance, pro_meth is the methylation level of the promoter region, body_meth is the methylation level of the gene body region, and Exp is the gene expression level. Rk(Var(pro_meth))＜301 The top 300 genes with the largest variance in their promoter regions are G Rk(Var(body_meth))＜301 G is the top 300 genes with the largest variance in the genome region. Rk(Var(Exp))＜301 The top 300 genes in terms of gene expression level variance.

[0057] 2) The obtained public gene G is initially saved, and the gene interaction network of the corresponding gene is found through the protein interaction network database (here, the STRING database is used as an example). A graph processing tool (here, the tool Cytoscape is used as an example) is used to filter out some scattered points in the gene interaction network. After integrating the omics data, a multi-omics graph data representation of various cancer samples is obtained;

[0058] For details, see Figure 3 , search the public gene G obtained in step 1) in the STRING database to obtain the gene interaction network of the corresponding gene.

[0059] Furthermore, after preliminary processing by the omics data processing method proposed in the present invention, i.e., step 1), three types of genomic data are obtained. Based on the three genomic data, the gene interaction network is integrated to generate a graph representation of each sample, thereby obtaining multi-omics graph data of various cancer samples. Here, edges with gene interaction confidence medium_Confidence greater than 0.4 are screened to obtain a preliminary sample gene interaction network skeleton. Then, based on the sample gene interaction network skeleton and the three genomic data obtained in step 1), the sample graph representation algorithm is used to integrate them to obtain a multi-omics graph data representation S of various cancer samples, i.e., a multi-omics cancer sample representation based on graph representation, S = {(G1, y1), (G2, y2), (G3, y3) ... (G N ,y N )}, where S represents a certain cancer type graph data representation set, and a certain cancer type graph data representation set S includes Sample1, Sample2, ..., SampleN, G i represents the i-th sample in the dataset, and yi represents the cancer subtype label corresponding to the i-th sample.

[0060] Among them, see Table 1, the specific process of the sample graph representation algorithm is:

[0061] The sample gene interaction network skeleton was reconstructed and represented as a two-dimensional list, where rows represent source nodes and columns represent target nodes. The values ​​in the list indicate whether an edge exists. Using this sample gene interaction network skeleton, we can determine whether an edge exists between two genes. If an edge exists, the corresponding node's adjacent node number is used as the feature of the current node; otherwise, the search continues until the entire two-dimensional list is traversed.

[0062] Then, for each node, the corresponding three genomic data sets are integrated and the integration results are used as the features of this node. This results in a graph data representation that integrates the gene interaction network skeleton and omics information, that is, a multi-omics cancer sample representation based on graph representation.

[0063] 3) Using the common gene G obtained in step 1) and the sample gene interaction network skeleton obtained in step 2), the cancer sample omics data of multiple samples can be integrated into one file, so that subsequent researchers can use it end-to-end (here taking typing prediction as an example) and reduce the overall coupling degree of the algorithm.

[0064] Example 1

[0065] In order to verify the effectiveness of the data processing method proposed in the present invention, the present invention proposes a cancer subtype classification method based on the cancer sample representation, the framework of which is as follows: Figure 3 The data source is multi-omics data from real cancer samples, and the network is from the STRING database. The constructed graph model is input into the graph neural network module. After passing through the convolution kernel graph pooling module, prediction is performed for downstream analysis, such as exploring tumor markers and biological function enrichment, which proves the effectiveness of this representation method.

[0066] Table 1 Sample graph representation algorithm

[0067]

[0068]

[0069] In order to verify the accuracy of the method proposed in the present invention, the present invention used breast cancer data in the TCGA public library database to verify the conclusion.

[0070] The verification process is as follows Figure 4As shown in the figure, first, the cancer omics data in the database is extracted, and then the STRING database is used to obtain the cancer gene interaction network skeleton. The sample graph model construction method proposed in the present invention is used to construct a graph of breast cancer samples. Then, a suitable graph neural network module is designed, which specifically includes a graph convolution module and a graph pooling module. The cancer sample graph is input for training, and finally the cancer subtype is predicted through the prediction generation layer. Some intermediate results in the network can be used to perform downstream analyses such as tumor marker prediction and biological function enrichment.

[0071] The present invention uses this framework to predict the molecular subtypes of breast cancer, and the prediction results are as follows: Figure 5 As shown in the figure, after 1,000 training rounds, the network's accuracy in predicting breast cancer subtypes can exceed 80%, which exceeds most of the existing baseline models.

[0072] This invention provides a process for processing high-dimensional cancer multi-omics data, enabling better screening of multi-omics data related to cancer development. In processing methylation data, this invention divides it into two data sets based on their differences in gene expression for processing and application. Previous cancer sample characterization methods, due to model establishment limitations, were unable to achieve cancer sample typing based on biological network modeling. This new cancer sample representation method proposed in this invention utilizes biological network modeling to model each sample as a graph representation.

[0073] Another embodiment of the present invention provides a multi-omics cancer sample representation system based on graph representation, comprising:

[0074] Data acquisition module, used to obtain gene public library data;

[0075] A data processing module is used to process the gene public library data to obtain gene expression data and methylation data of cancer samples with variance greater than a threshold;

[0076] The public gene acquisition module is used to process the gene expression data and methylation data of cancer samples with variance greater than a threshold to obtain multi-omics data of cancer samples. The multi-omics data of cancer samples are intersected to obtain public genes.

[0077] The sample gene interaction network skeleton acquisition module is used to obtain the gene interaction network of the corresponding gene based on the public gene through the protein interaction network database, filter out some scattered points in the gene interaction network and integrate them to obtain the sample gene interaction network skeleton;

[0078] The integration module is used to integrate the multi-omics data of cancer samples with the sample gene interaction network skeleton representation using a sample graph representation algorithm to obtain a multi-omics cancer sample representation based on graph representation.

[0079] Another embodiment of the present invention provides an electronic device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the graph-representation-based multi-omics cancer sample representation method.

[0080] Another embodiment of the present invention provides a computer-readable storage medium storing a computer program. When the computer program is executed by a processor, a multi-omics cancer sample representation method based on graph representation is implemented.

[0081] The above description is merely a description of the preferred embodiment of the present invention and is not to be construed as limiting the claims. The present invention is not limited to the above embodiment, and variations in the specific structure are permitted. Any variations made within the scope of the independent claims of the present invention are also within the scope of protection of the present invention.

[0082] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which this invention pertains. The terms used herein in the specification of the present invention are for the purpose of describing specific embodiments only and are not intended to limit the present invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items.

Claims

1. A multi-omics cancer sample representation method based on graph representation, characterized in that: The following steps are involved: Access to public gene bank data; Processing the public gene library data to obtain gene expression data and methylation data of cancer samples with variance greater than the threshold; After processing the gene expression data and methylation data of cancer samples with variance greater than the threshold, multi-omics data of cancer samples are obtained. The multi-omics data of cancer samples include sample expression matrix, promoter region methylation matrix and gene body region methylation matrix. The multi-omics data of cancer samples are intersected to obtain common genes. The specific process includes: The molecular omics data of samples obtained from the public cancer sample database Cancer Genome Atlas were used as the data source. The omics data pre-processed in the UCSC Xena database were used to extract the gene expression data files and methylation data files HM450 files of the samples. The gene expression data files of the samples were subjected to variance filtering to obtain the sample expression matrix. The methylation data files HM450 files and the CHAMP probe-gene name package in the R language were subjected to variance filtering to obtain the promoter region methylation matrix and the gene body region methylation matrix. The columns of the sample expression matrix, promoter region methylation matrix and gene body region methylation matrix represent one gene and the rows represent one sample. Then, according to formula (3), the top 300 genes with the highest variance in the three matrices were selected. （3） Where G is the common gene, Rk is the variance ranking of genomic data, Var is the variance, pro_meth is the methylation level of the promoter region, body_meth is the methylation level of the gene body region, and Exp is the gene expression level; The top 300 genes with the largest variance in their promoter regions are: The top 300 genes with the largest variance in the genome region. The top 300 genes in terms of gene expression level variance; Among them, the methylation data for genes The methylation level of the DNA methylation region that inhibits expression was calculated by the following formula: （1） in, Indicates gene The methylation level of the DNA methylation region that inhibits expression is the methylation level of the promoter region. Indicates the methylation status of the upstream region from the transcription start site of the gene. Indicates the methylation status of the first exon region of the gene. Indicates the methylation status of the 5' untranslated region of the gene. Methylation status of the upstream region from the transcription start site of the gene; Methylation data for genes The methylation level of the DNA methylation region that promotes expression is calculated by the following formula: （2） Where, Indicates gene The methylation level of the DNA methylation region that promotes expression, Indicates gene The methylation level of the gene body region in the region that promotes expression. Gene body represents the gene body region, i represents the gene number, and Median represents the median. Based on public genes, the gene interaction network of the corresponding genes is obtained through the protein interaction network database. After filtering out some scattered points in the gene interaction network and integrating them, the sample gene interaction network skeleton is obtained; the multi-omics data of cancer samples and the sample gene interaction network skeleton representation are integrated using the sample graph representation algorithm to obtain a multi-omics cancer sample representation based on graph representation. The specific process is to preliminarily save the public gene G, find the gene interaction network of the corresponding gene through the protein interaction network database, use the graph processing tool to filter out some scattered points in the gene interaction network, integrate the multi-omics data of cancer samples with the gene interaction network to generate a graph representation of each sample, thereby obtaining multi-omics graph data of various cancer samples, and screen the edges with gene interaction confidence greater than 0.4 to obtain a preliminary sample gene interaction network skeleton. Then, based on the sample gene interaction network skeleton and the three genomics data, the sample graph representation algorithm is used to integrate them to obtain multi-omics graph data representation of various cancer samples; The sample graph representation algorithm is as follows: the sample gene interaction network skeleton is reconstructed and represented as a two-dimensional list, where rows represent source nodes, columns represent target nodes, and the values ​​in the list represent whether there is an edge. Using the sample gene interaction network skeleton, it is possible to determine whether there is an edge between two genes; if there is an edge, the adjacent node number of the corresponding node is used as the feature of the node; otherwise, the search will continue until the entire two-dimensional list is traversed; Then, for each node, the corresponding three genomic data are integrated, and the integration results are used as the features of this node to obtain a multi-omics cancer sample representation based on graph representation.

2. The multi-omics cancer sample representation method based on graph representation according to claim 1, characterized in that: The multi-omics cancer sample representation S based on graph representation is: Among them, S represents a certain cancer type graph data set, represents the i-th sample in the dataset, Represents the cancer subtype label corresponding to the i-th sample.

3. A multi-omics cancer sample representation system based on graph representation, characterized by: include: Access to public gene bank data; Processing the public gene library data to obtain gene expression data and methylation data of cancer samples with variance greater than the threshold; After processing the gene expression data and methylation data of cancer samples with variance greater than the threshold, multi-omics data of cancer samples are obtained. The multi-omics data of cancer samples include sample expression matrix, promoter region methylation matrix and gene body region methylation matrix. The multi-omics data of cancer samples are intersected to obtain common genes. The specific process includes: The molecular omics data of samples obtained from the public cancer sample database Cancer Genome Atlas were used as the data source. The omics data pre-processed in the UCSC Xena database were used to extract the gene expression data files and methylation data files HM450 files of the samples. The gene expression data files of the samples were subjected to variance filtering to obtain the sample expression matrix. The methylation data files HM450 files and the CHAMP probe-gene name package in the R language were subjected to variance filtering to obtain the promoter region methylation matrix and the gene body region methylation matrix. The columns of the sample expression matrix, promoter region methylation matrix and gene body region methylation matrix represent one gene and the rows represent one sample. Then, according to formula (3), the top 300 genes with the highest variance in the three matrices were selected. （3） Where G is the common gene, Rk is the variance ranking of genomic data, Var is the variance, pro_meth is the methylation level of the promoter region, body_meth is the methylation level of the gene body region, and Exp is the gene expression level; The top 300 genes with the largest variance in their promoter regions are: The top 300 genes with the largest variance in the genome region. The top 300 genes in terms of gene expression level variance; Among them, the methylation data for genes The methylation level of the DNA methylation region that inhibits expression was calculated by the following formula: （1） in, Indicates gene The methylation level of the DNA methylation region that inhibits expression is the methylation level of the promoter region. Indicates the methylation status of the upstream region from the transcription start site of the gene. Indicates the methylation status of the first exon region of the gene. Indicates the methylation status of the 5' untranslated region of the gene. Methylation status of the upstream region from the transcription start site of the gene; Methylation data for genes The methylation level of the DNA methylation region that promotes expression is calculated by the following formula: （2） Where, Indicates gene The methylation level of the DNA methylation region that promotes expression, Indicates gene The methylation level of the gene body region in the region that promotes expression. Gene body represents the gene body region, i represents the gene number, and Median represents the median. Based on public genes, the gene interaction network of the corresponding genes is obtained through the protein interaction network database. After filtering out some scattered points in the gene interaction network and integrating them, the sample gene interaction network skeleton is obtained; the multi-omics data of cancer samples and the sample gene interaction network skeleton representation are integrated using the sample graph representation algorithm to obtain a multi-omics cancer sample representation based on graph representation. The specific process is to preliminarily save the public gene G, find the gene interaction network of the corresponding gene through the protein interaction network database, use the graph processing tool to filter out some scattered points in the gene interaction network, integrate the multi-omics data of cancer samples with the gene interaction network to generate a graph representation of each sample, thereby obtaining multi-omics graph data of various cancer samples, and screen the edges with gene interaction confidence greater than 0.4 to obtain a preliminary sample gene interaction network skeleton. Then, based on the sample gene interaction network skeleton and the three genomics data, the sample graph representation algorithm is used to integrate them to obtain multi-omics graph data representation of various cancer samples; The sample graph representation algorithm is as follows: the sample gene interaction network skeleton is reconstructed and represented as a two-dimensional list, where rows represent source nodes, columns represent target nodes, and the values ​​in the list represent whether there is an edge. Using the sample gene interaction network skeleton, it is possible to determine whether there is an edge between two genes; if there is an edge, the adjacent node number of the corresponding node is used as the feature of the node; otherwise, the search will continue until the entire two-dimensional list is traversed; Then, for each node, the corresponding three genomic data are integrated, and the integration results are used as the features of this node to obtain a multi-omics cancer sample representation based on graph representation.

4. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: When the processor executes the computer program, the multi-omics cancer sample representation method based on graph representation according to any one of claims 1 to 2 is implemented.

5. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the multi-omics cancer sample representation method based on graph representation according to any one of claims 1 to 2 is implemented.

Images