A ceRNA network-based cancer gene mining analysis method

Abstract

The application discloses a cancer gene mining analysis method based on a ceRNA network, and steps are as follows: gene expression data and pathological images of patients suffering from target cancer and patients without diseases are acquired; relevant pathological images are processed, nuclear level features of cell nuclei cut by the pathological images are extracted, and the nuclear level features are taken as WSI features; a ceRNA network reflecting connection relationships between different RNAs is constructed according to the gene expression data; a sparse matrix of different RNA pairs is established, and Pearson correlation coefficient solving is performed on the WSI features and mRNA data; JSNMCNMF algorithm is used to integrate four parts of WSI, mRNA, miRNA and lncRNA data, and the relationship between mRNA and WSI is obtained; and KJMCNMF algorithm is used to acquire hidden connection relationships in the ceRNA network. The method of the application firstly constructs the relationship between genes and pathological images, and provides a new direction for cancer research.

Description

Technical Field

[0001] This invention relates to the field of bioinformatics technology, and specifically to a method for cancer gene mining and analysis based on ceRNA networks. Background Technology

[0002] Cancer is one of the leading causes of death worldwide, with lung cancer, stomach cancer, liver cancer, and breast cancer being the most common. It can occur in any part of the body and is also known as a malignant tumor. It is caused by the loss of normal cellular regulation, resulting in the excessive proliferation of abnormal cells, which are often invasive and metastasize to other parts of the body via the circulatory and lymphatic systems, causing serious damage. At the molecular level, cancer causes mutations in the genome of a cell, including point mutations, insertion mutations, deletion mutations, and chromosomal translocations. Externally, it can lead to abnormal cell or tissue growth, leaving symptoms deep within the patient's skin or tissues. In recent years, with the continuous improvement of modern medical technology, in addition to surgery, biotherapy, chemotherapy, and radiotherapy, targeted therapy and personalized medicine have become emerging treatment methods. These methods identify mutation sites at the gene level that differ from normal cells and use drugs to combine with the cancerous sites, thereby causing tumor cells to become ineffective. However, the cost of general gene sequencing is very expensive, and how to effectively and cost-effectively identify mutated genes has become a major focus of bioinformatics in recent years.

[0003] The ceRNA network is a competitive endogenous hypothesis that helps biologists understand the interactions and mechanisms between RNAs. Within the ceRNA network are miRNA response elements (MREs), which have been found not only on mRNA but also on other types of RNA such as lncRNAs. Therefore, miRNAs can bind to different RNAs through MREs. When two RNAs compete for the same miRNA molecule, changes in the expression of one gene due to cancer can lead to upregulation or downregulation of the other gene, reflecting the internal relationships. lncRNAs are considered to be associated with many human diseases and cancer. Analysis of the ceRNA network can identify significantly relevant mutation sites in cancer and their development during pathogenesis. However, several key issues remain to be addressed systematically, including how to select and identify interacting gene pairs and predict new relationships.

[0004] Furthermore, doctors use warp-scan (WSI) images to qualitatively classify and clinically analyze tumors. However, analyzing thousands of patient pathology images is a very tedious process, especially for younger, less experienced doctors. Therefore, if we could obtain information about the underlying genetic changes and their mechanisms through WSI images, it would be of great significance for patient treatment and prognosis, as well as for disease research.

[0005] Therefore, developing a method that can analyze cancer genes based on WSI images is of great practical significance. Summary of the Invention

[0006] Due to the aforementioned deficiencies in existing technologies, this invention provides a method for analyzing cancer genes based on WSI images, overcoming the limitation of existing technologies in failing to achieve this process. Specifically, it obtains the cell nucleus information of each patient through a multi-layer, multi-threshold histological image cell nucleus segmentation method, and obtains nuclear-level features and image features through feature extraction; then, it constructs a ceRNA network based on a database, and finally integrates WSI images and three types of RNA data using the JSNMCNMF method to obtain their connectivity relationships; and uses KJMCNMF to obtain the hidden connectivity relationships in the high-dimensional ceRNA network.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A method for cancer gene mining and analysis based on ceRNA networks includes the following steps:

[0009] (1) Obtain gene expression data and pathological images corresponding to the target cancer, wherein the gene expression data and pathological images corresponding to the target cancer include gene expression data and pathological images of patients with the target cancer and patients without the cancer, and the gene expression data includes gene expression data of miRNA, mRNA and lncRNA.

[0010] (2) Process the relevant pathological images and extract the nuclear-level features of the cell nuclei cut from the pathological images as WSI features;

[0011] (3) Construct a ceRNA network reflecting the connection relationships between different RNAs based on gene expression data;

[0012] (4) Establish a sparse matrix of pairings of different RNAs, and solve the Pearson correlation coefficient between WSI features and mRNA data;

[0013] (5) The JSNMCNMF algorithm was used to integrate the four parts of data: WSI, mRNA, miRNA and lncRNA, to obtain the relationship between mRNA and WSI; the KJMCNMF algorithm was used to obtain the hidden connection relationship in the ceRNA network.

[0014] This invention aims to utilize gene expression data from cancer patients, combining ceRNA networks with pathological images to fully extract features from WSI images. A Joint Sparse Network-Regularization Multi-Constrained Non-negative Matrix Factorization (JSNMCNMF) algorithm is employed to obtain the relationship between WSI and genes. Furthermore, a Kernel Joint Multi-Constrained Non-negative Matrix Factorization (KJMCNMF) algorithm is used to analyze and predict the regulatory relationships of various RNAs within the ceRNA network at a higher dimension. This allows for the identification of cancer-related genes and potential biomarkers, enabling researchers to conduct further investigations without wasting resources on comprehensive gene sequencing.

[0015] As a preferred technical solution:

[0016] The cancer gene mining and analysis method based on ceRNA networks described above, specifically step (2) is as follows:

[0017] (2.1) Cut the relevant pathological images and define the side length of each small block as pst. Store the coordinates of the upper left corner of each small block in the matrix xy. For each small block, the sum of the pixels with a pixel value greater than 210 must be greater than the square of the side length of the small block divided by 2. A pixel value greater than 210 means that the pixel is located in the stained tissue area, that is, the small block must contain cell tissue rather than the surrounding blank area. Remove small blocks that exceed the side length of the image.

[0018] (2.2) Perform color deconvolution and color normalization operations on each obtained small block;

[0019] Color deconvolution and color normalization operations are beneficial for enhancing cell nucleus segmentation by utilizing different staining information in the image. Specifically, we define the image vectorization as J, J = [i1, i2, i3, ..., in], where i represents a pixel and contains three data i[r, g, b], representing three color channels, calculated using the following formula:

[0020]

[0021] Where M is the default color deconvolution matrix proposed by Ruifork and Johnston, and is finally reconstructed into an image-form matrix DCH. We select the R channel and normalize it to the data range of 0 to 1.

[0022]

[0023] (2.3) Perform a protective erosion operation on the processed image, using a disk with a radius of 3 as a template, and erode the pixels except for the cell nucleus;

[0024] (2.4) A multi-level thresholding operation is adopted, with the number of levels set to 10. The multithresh function is used to obtain the pixel threshold of each level of the image according to the number of levels to be segmented. The smallest threshold of each level is selected and sorted in descending order. For each level, a binarization operation is performed according to the threshold, and the white area represents the cell nucleus. The cell nucleus is filled and the black dots are removed. Then, an opening operation is performed to separate the cell nuclei with connected boundaries. Cell nuclei with too small area are removed. Finally, the obtained image is compared with the image obtained from the previous level. Because it is in descending order, the ratio of cell nucleus area to number in the previous level image is greater than that in the current level image. The areas in the previous level image with cell nucleus connected components greater than 2 are replaced with the current area to improve the segmentation accuracy.

[0025] (2.5) Perform boundary processing. If the number of connected cell nuclei after expansion does not exceed 2, then expand the cell nuclei and finally smooth the surface.

[0026] (2.6) For the processed small image, calculate the area of ​​each cell nucleus (representing cell nucleus size), the length of the long axis of the cell nucleus, the length of the short axis of the cell nucleus, the ratio of the long axis to the short axis of the cell nucleus, the farthest distance, the nearest distance and the average distance between the cell nucleus and its neighboring cell nucleus (representing cell nucleus density), and the average value of the three-channel color of the cell nucleus. The first 7 features can be obtained using the regionprops function, which can obtain the centroid of each cell nucleus and calculate its distance based on the centroid.

[0027] (2.7) The data obtained above are kernel-level features. The kernel-level feature data are clustered. The K-means algorithm is used to define 10 centroids. Each cell kernel feature is assigned to the centroid closest to it. Finally, the frequency is obtained by dividing by the total number and generating 10 histograms. In addition, 5 statistical distributions are performed on each kernel-level feature, namely mean μ, standard deviation σ, skewness s, kurtosis k and entropy Ent.

[0028] The formula is as follows:

[0029]

[0030]

[0031]

[0032] Where x represents the value of each feature, and P(x) is the frequency of that value in the feature, resulting in a total of 150 features.

[0033] The cancer gene mining and analysis method based on ceRNA networks described above, specifically step (3) is as follows:

[0034] (3.1) The gene expression data is summarized into a matrix of gene IDs and behavioral sample IDs;

[0035] (3.2) Convert gene IDs to gene names for display;

[0036] (3.3) Perform differential analysis on various RNA data to obtain differentially expressed genes DEmRNA, DEmiRNA, and DElncRNA in the three types of RNA data;

[0037] (3.4) Obtain the pairing information related to DElncRNA and find the data related to DEmiRNA from the pairing information to obtain lncRNA-miRNA pairs;

[0038] (3.5) Annotate the paired miRNAs, and then retrieve the target genes of the miRNAs from the database to obtain the target mRNAs and obtain miRNA-mRNA pairs;

[0039] (3.6) Draw a ceRNA network based on miRNA-mRNA pairs and lncRNA-miRNA pairs.

[0040] In the above-described method for cancer gene mining and analysis based on ceRNA networks, step (3.3) involves differential analysis using Deseq2.

[0041] Suppose we have a gene expression matrix X. First, we standardize X:

[0042]

[0043] Where Xaverage represents the average of each row, median is the median of each column, and n is the total number of samples. Then, the IfcSE standard error, SEM, Foldchange, and P-value are calculated.

[0044]

[0045]

[0046] in The average value for all patients with each disease in each feature. SD represents the standard deviation of all normal samples for each feature. We selected differentially expressed genes with p-values ​​less than 0.05 and FC absolute values ​​greater than 2. Using this method, we obtained differentially expressed genes DEmRNA, DEmiRNA, and DElncRNA in the three RNA datasets.

[0047] As described above, in a cancer gene mining and analysis method based on ceRNA networks, the annotation in step (3.5) refers to labeling the paired miRNA data with 3p and 5p according to the Starbase database;

[0048] The search was performed from the miRDB, miRTarBase, and TargetScan databases.

[0049] As described above, in a cancer gene mining and analysis method based on ceRNA networks, step (3.6) involves drawing the ceRNA network using Cytoscape software, where a positive FC value indicates upregulation, and a negative FC value indicates downregulation.

[0050] In the cancer gene mining and analysis method based on ceRNA networks described above, step (4) specifically includes:

[0051] (4.1) For each pairing, construct the connected sparse matrix between the two, namely the mRNA-miRNA sparse relation matrix and the lncRNA-miRNA sparse relation matrix, the values ​​of which are calculated by the following formula:

[0052]

[0053] Where a and b represent the FC values ​​of two interconnected relationships, resulting in a 3347*168 mRNA-miRNA sparse relation matrix A and a 2282*168 lncRNA-miRNA sparse relation matrix B.

[0054] (4.2) The relationship matrix between mRNA and WSI is calculated using Pearson correlation as a metric, and the formula is as follows:

[0055]

[0056] Where X1 represents the mRNA feature matrix, Represented as Expectations Represented as The variance of X2 is represented by the WSI matrix. Represented as Expectations Represented as The variance of the mRNA-WSI is calculated, and 1≤x≤3347, 1≤y≤150, resulting in a 3347*150 mRNA-WSI relation matrix C.

[0057] In the cancer gene mining and analysis method based on ceRNA networks described above, step (5) specifically includes:

[0058] (5.1) Construct the interaction adjacency matrices of mRNA-miRNA, lncRNA-miRNA, and WSI-mRNA respectively;

[0059] The interaction formula is shown below:

[0060]

[0061]

[0062]

[0063] Where a, b, and c are elements of the adjacency matrix, h i For the i-th row of H, h j The j-th column is represented by H, where H1, H2, H3, and H4 are the WSI, mRNA, miRNA, and lncRNA data, respectively.

[0064] (5.2) The data is clustered using the JSNMCNMF method, and its objective function is:

[0065]

[0066] Where α is the parameter controlling the orthogonality of the coefficient HI, λ is the weight of the constraint relationship, r1>0 is used to limit the growth of W, and r2>0 is used to enhance sparsity; we set a value K, and use the formula to decompose each matrix X into an n*K W matrix and a K*feature H matrix, where n is the number of samples and feature is the feature.

[0067] (5.3) For each module, i.e., each row of the H matrix, select features based on the Z-score:

[0068]

[0069] Where X ij For each element in H, μ is the mean of the i-th row in H, and σ is the standard deviation of the i-th row in H. If it is greater than the threshold T, then this feature is selected and added to the module. The features of each module are analyzed to obtain the relationship between WSI and other RNAs.

[0070] (5.4) For miRNA, mRNA, and lncRNA, X for each type of data I Mapped to higher-dimensional space middle;

[0071] Using kernel functions

[0072]

[0073] Let represent the distance between two sets of data, where I and J represent the two different sets of data, i represents the row of the matrix, j represents the column of the matrix, n is the number of patients in the data, and σ is used to set the control range of the kernel function. The KJMCNMF method is used to perform high-dimensional clustering on the data, with the objective function being:

[0074]

[0075] Here, w controls the similarity between the XQ matrices. We let the coefficient matrix W = XQ, which is understood as a convex combination of certain eigenvalues ​​of the original data X, i.e., A is similar to the coefficient matrix W. Each type of data X is decomposed into XQH. For the jointness of the data, the XQ matrices of different combinations should differ as little as possible. The H matrix is ​​calculated, and using the same steps as in 5.3 above, the features of each module are analyzed, and Pearson correlation matrices are built in batches to analyze latent biomarkers and latent connections.

[0076] The above technical solution is only one feasible technical solution of the present invention. The scope of protection of the present invention is not limited thereto. Those skilled in the art can reasonably adjust the specific design according to actual needs.

[0077] The above invention has the following advantages or beneficial effects:

[0078] The cancer gene mining and analysis method based on ceRNA networks of this invention connects WSI data with RNA molecular interaction networks, which can effectively discover potential correlation patterns between pathological images and gene expression data, as well as the relationship between images and the internal mechanism of operation. This invention uses the KJMCNMF algorithm to observe the connection relationship between ceRNA networks at a high dimension. The use of nonlinear NMF method can improve the performance of the results and is closer to the complex data structure in reality, showing good application prospects. Attached Figure Description

[0079] The invention, its features, shape, and advantages will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings. Like reference numerals denote like parts throughout the drawings. The drawings are not drawn to scale; their focus is on illustrating the gist of the invention.

[0080] Figure 1This is a schematic flowchart of the cancer gene mining and analysis method based on ceRNA network of the present invention. Detailed Implementation

[0081] The structure of the present invention will be further described below with reference to the accompanying drawings and specific embodiments, but this is not intended to limit the present invention.

[0082] Example 1

[0083] A cancer gene mining and analysis method based on ceRNA networks, such as Figure 1 As shown, it includes the following steps:

[0084] Step 1: Transcriptome data (including lncRNA and mRNA) from 551 LUSC patients were downloaded from the TCIA database, containing data from 502 patients and 49 normal individuals. miRNA data from 523 LUSC patients were also downloaded, containing data from 478 patients and 45 normal individuals. 1100 WSI images, including both normal and diseased cases, were downloaded from the GDC website. Patients appearing in each dataset were filtered based on the available data, resulting in a total of 411 patient data points.

[0085] Step 2.1: Traverse the 411 WSI images, segment each image into small blocks, and for each small block, use a multi-layer, multi-threshold histological image cell nucleus segmentation method. The steps are: color deconvolution → image reconstruction → multi-level threshold determination → image binarization → image filling → opening operation → smoothing.

[0086] Step 2.2: For each cell nucleus in the obtained image, calculate its area; the average value of the three image channels; the length of the cell nucleus's major axis, minor axis, and the ratio of its major and minor axes; and the shortest, longest, and average distances between adjacent cell nuclei. For these 10 features, calculate their mean, standard deviation, skewness, kurtosis, and entropy. Skewness is a measure of the asymmetry of the data distribution around the sample mean; kurtosis is a measure of the distribution's tendency towards discrete values; and entropy is a measure of randomness. Taking the cell nucleus feature of the ratio of its major and minor axes as an example, its corresponding histogram features are ratio_bin1 to ratio_bin10, corresponding to five distribution statistical representations: ratio_mean, ratio_std, ratio_skewness, ratio_kurtosis, and ratio_entropy. Other nucleus-level features follow the same pattern, totaling 150 features. Among them, histogram ratio_bin1 represents a round cell nucleus, and ratio_bin10 represents a very long and thin cell nucleus, such as a lung squamous cell. Finally, a 411*150 WSI feature matrix is ​​obtained.

[0087] Step 3.1: First, the data downloaded from the website is scattered. We need the gene expression data for each patient, summarized into a matrix with gene IDs and behavioral sample IDs. Integrate the gene data by extracting the data for each subject from the compressed package and merging them into a 60660*551 matrix containing lncRNA, mRNA, and other RNA types. Use the Ensembl database website to convert the gene IDs into gene names for display. Based on the gene names, divide the matrix into a 19763*551 mRNA matrix X1 and a 14321*551 lncRNA matrix X3. Similarly, obtain a 1881*523 miRNA matrix X2.

[0088] Step 3.2: Differential analysis was performed using Deseq2. We selected genes with an absolute FC value greater than 2 and a p-value less than 0.05, obtaining data for 3347 differentially expressed mRNAs, 168 differentially expressed miRNAs, and 2282 differentially expressed lncRNAs.

[0089] Step 3.3: Use the mircode database to find all pairing information related to DElncRNA. In the pairing information, find the data related to DEmiRNA, and obtain a total of 635 lncRNA-miRNA pairs.

[0090] Step 3.4: The paired miRNA data is labeled with 3p and 5p according to the starbase database. For the labeled miRNA data, the target genes of the miRNA, i.e. the target mRNA, are retrieved from the miRDB, miRTarBase, and TargetScan databases, resulting in 509 miRNA-mRNA pairs.

[0091] Step 3.5: Based on the obtained pairings, use Cytoscape software to draw the ceRNA network. Positive FC values ​​are marked as upregulated in red, while negative FC values ​​are marked as downregulated in green.

[0092] Step 4.1: For each pairing, construct a connected sparse matrix between the two pairs, the values ​​of which are calculated using the following formula:

[0093]

[0094] Where a and b represent the FC values ​​of two interconnected relationships, we obtain a 3347*168 mRNA-miRNA sparse relation matrix A and a 2282*168 lncRNA-miRNA sparse relation matrix B.

[0095] Step 4.2: Calculate the relationship matrix between mRNA and WSI, using Pearson correlation as the metric. The formula is as follows:

[0096]

[0097] Where X1 represents the mRNA feature matrix, Represented as Expectations Represented as The variance of X2 is represented by the WSI matrix. Represented as Expectations Represented as The variance of the mRNA-WSI is calculated, and 1≤x≤3347, 1≤y≤150, resulting in a 3347*150 mRNA-WSI relation matrix C.

[0098] Step 5: Using the JSNMCNMF and KJMCNMF methods, cluster the data to obtain different modules in the H matrix. Then, perform Z-score scoring and analysis on these modules to obtain their hidden biomarkers and hidden connections. The solution to the H matrix is ​​shown below:

[0099]

[0100]

[0101]

[0102]

[0103]

[0104] The above equation is the solution for the JSNMCNMF method, and the below equation is the solution for the KJMCNMF method.

[0105]

[0106]

[0107]

[0108] Those skilled in the art should understand that variations can be implemented by combining existing technology with the above embodiments, which will not be elaborated here. Such variations do not affect the essence of the present invention, and will not be elaborated here either.

[0109] The preferred embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the specific embodiments described above, and the devices and structures not described in detail should be understood as being implemented in a conventional manner in the art. Any person skilled in the art can make many possible variations and modifications to the technical solutions of the present invention using the methods and techniques disclosed above, or modify them into equivalent embodiments with equivalent changes, without departing from the scope of the present invention. This does not affect the essential content of the present invention. Therefore, any simple modifications, equivalent changes, and modifications made to the above embodiments based on the technical essence of the present invention without departing from the content of the present invention's technical solutions still fall within the protection scope of the present invention.

Claims

1. A method for cancer gene mining and analysis based on ceRNA networks, characterized in that: Includes the following steps: (1) Obtain gene expression data and pathological images corresponding to the target cancer, wherein the gene expression data and pathological images corresponding to the target cancer include gene expression data and pathological images of patients with the target cancer and patients without the cancer, and the gene expression data includes gene expression data of miRNA, mRNA and lncRNA; (2) Process the relevant pathological images and extract the nuclear-level features of the cell nuclei cut from the pathological images as WSI features; (3) Construct a ceRNA network reflecting the connection relationships between different RNAs based on gene expression data; (4) Establish a sparse matrix of pairings of different RNAs, and solve the Pearson correlation coefficient between WSI features and mRNA data; (5) The JSNMCNMF algorithm was used to integrate the four parts of data: WSI, mRNA, miRNA and lncRNA, to obtain the relationship between mRNA and WSI; the KJMCNMF algorithm was used to obtain the hidden connection relationship in the ceRNA network. Step (2) is as follows: (2.1) Cut the relevant pathological images and define the side length of each small block as pst. Store the coordinates of the upper left corner of each small block in the matrix xy. For each small block, the sum of the pixels with a pixel value greater than 210 must be greater than the square of the side length of the small block divided by 2. A pixel value greater than 210 means that the pixel is located in the stained tissue area. Remove small blocks that exceed the side length of the image. (2.2) Perform color deconvolution and color normalization operations on each obtained small block; (2.3) Perform a protective erosion operation on the processed image, using a disk with a radius of 3 as a template, and erode the pixels except for the cell nucleus; (2.4) A multi-level thresholding operation is adopted. The multithresh function is used to obtain the pixel threshold of each level of the image according to the number of levels to be segmented. The minimum threshold of each level is selected and sorted in descending order. For each level, a binarization operation is performed according to the threshold, and the white area represents the cell nucleus. The cell nucleus is filled and the black dots are removed. Then, an opening operation is performed to separate the cell nuclei with connected boundaries. Cell nuclei with too small area are removed. Finally, the obtained image is compared with the image obtained in the previous level. The areas with cell nucleus connected components greater than 2 in the previous level image are replaced with the current area for display. (2.5) Perform boundary processing. If the number of connected cell nuclei after expansion does not exceed 2, then expand the cell nuclei and finally smooth the surface. (2.6) For the processed small image, count the area of ​​each cell nucleus, the length of the long axis of the cell nucleus, the length of the short axis of the cell nucleus, the ratio of the long axis to the short axis of the cell nucleus, the farthest distance, the closest distance and the average distance between the cell nucleus and the adjacent cell nucleus, and the average value of the three-channel color of the cell nucleus; (2.7) The data obtained above are kernel-level features. The kernel-level feature data are clustered. The K-means algorithm is used to define 10 centroids. Each cell kernel feature is assigned to the centroid closest to it. Finally, the frequency is obtained by dividing by the total number and generating 10 histograms. In addition, 5 statistical distributions are performed on each kernel-level feature, namely the mean μ, standard deviation σ, skewness s, kurtosis k and entropy Ent.

2. The method for cancer gene mining and analysis based on ceRNA networks according to claim 1, characterized in that, The specific steps (3) are as follows: (3.1) The gene expression data is summarized into a matrix of gene IDs and behavioral sample IDs; (3.2) Convert gene IDs to gene names for display; (3.3) Perform differential analysis on various RNA data to obtain differentially expressed genes DEmRNA, DEmiRNA, and DElncRNA in the three types of RNA data; (3.4) Obtain the pairing information related to DElncRNA and find the data related to DEmiRNA from the pairing information, that is, obtain the lncRNA-miRNA pair; (3.5) Annotate the paired miRNAs, and then retrieve the target genes of the miRNAs from the database to obtain the target mRNAs and obtain miRNA-mRNA pairs; (3.6) Draw a ceRNA network based on miRNA-mRNA pairs and lncRNA-miRNA pairs.

3. The method for cancer gene mining and analysis based on ceRNA networks according to claim 2, characterized in that, Step (3.3) involves performing differential analysis using Deseq2.

4. The method for cancer gene mining and analysis based on ceRNA networks according to claim 2, characterized in that, The annotation mentioned in step (3.5) refers to the 3p and 5p annotation of the paired miRNA data according to the starbase database; The search was performed from the miRDB, miRTarBase, and TargetScan databases.

5. The method for cancer gene mining and analysis based on ceRNA networks according to claim 2, characterized in that, Step (3.6) involves using Cytoscape software to draw the ceRNA network.

6. The method for cancer gene mining and analysis based on ceRNA networks according to claim 1, characterized in that, Step (4) specifically involves: (4.1) For each pairing, establish the connected sparse matrix between the two, namely the mRNA-miRNA sparse relation matrix and the lncRNA-miRNA sparse relation matrix; (4.2) The relationship matrix between mRNA and WSI was calculated using Pearson correlation as a metric.

7. The method for cancer gene mining and analysis based on ceRNA networks according to claim 1, characterized in that, Step (5) specifically involves: (5.1) Construct the interaction adjacency matrices of mRNA-miRNA, lncRNA-miRNA, and WSI-mRNA respectively; (5.2) Cluster the data using the JSNMCNMF method; (5.3) For each module, select features based on the Z-score; (5.4) For miRNA, mRNA, and lncRNA, X for each type of data I Mapped to the high-dimensional space Ф(X) I In the process, the KJMCNMF method is then used to perform high-dimensional clustering on the data.

Images