Gene co-expression network identification method and system based on graph convolutional neural network

Abstract

The invention relates to a gene co-expression network identification method and system based on a graph convolutional neural network. The method comprises the following steps: S1, data preprocessing; s2, constructing a gene network diagram based on gene expression correlation; s3, node information is captured through a GraphSAGE model of the graph convolutional neural network, and features are extracted; s4, performing clustering analysis by using a Gaussian mixture model; and S5, dividing gene modules according to a clustering result. The method can help researchers to more effectively extract valuable information from massive gene expression data, deeply understand a gene regulatory network, find new gene functions and regulatory mechanisms, and provide important guidance for disease treatment and prevention. And meanwhile, the method also has good expandability and flexibility, and can adapt to analysis requirements of gene expression data of different scales and types.

Description

Technical Field

[0001] The present invention belongs to the field of bioinformatics, and in particular relates to a gene co-expression network identification method and system based on graph convolutional neural network. Background Art

[0002] In recent years, with the rapid development of bioinformatics, the analysis of gene expression data has become one of the important means to reveal the mysteries of life sciences. Gene expression data reflects the expression level of genes in organisms under different conditions, which is of great significance for understanding gene function, regulatory mechanism and disease mechanism. However, faced with the huge complexity of gene expression data, how to extract valuable information from massive data is still one of the challenges facing the current field of bioinformatics. The widely used gene co-expression network analysis methods are mainly based on biostatistics techniques. These methods face problems such as high computational complexity and insufficient precision when processing large-scale genomic data. Summary of the invention

[0003] In order to address the shortcomings of the prior art, the present invention introduces the graph convolutional network GraphSAGE in the field of deep learning to perform neighbor sampling and aggregation to achieve feature selection, and on this basis, identifies gene co-expression modules. It is expected to reveal the regulatory relationship between genes more accurately and efficiently, and then discover some gene functions that have not yet been recognized, which has important guiding significance for the treatment and prevention of diseases.

[0004] The technical solution of the present invention is:

[0005] A gene co-expression network identification method based on graph convolutional neural network comprises the following steps:

[0006] S1. Data preprocessing to reduce data noise and scale differences.

[0007] S2. Constructing a gene network graph based on gene expression correlation: Based on the preprocessed gene expression data, the gene network graph is constructed by calculating the correlation between genes (such as the Pearson correlation coefficient). In the graph, genes are used as nodes, and the correlation between genes is used as the weight of the edge.

[0008] S3. Capture node information and extract features through the graph convolutional neural network GraphSAGE model; Use the graph convolutional neural network GraphSAGE model to capture and extract features from node information in the gene network graph, and combine the sampler to sample and aggregate the neighbors of the node. GraphSAGE can learn the characteristics of the node and its neighboring nodes, thereby obtaining the complex regulatory relationship between genes.

[0009] S4. Cluster analysis using Gaussian mixture model: Cluster analysis of features extracted by graph convolutional network was performed using Gaussian mixture model (GMM). GMM can fit the complex distribution of data and identify different gene modules. The clustering results divide genes into different co-expression modules.

[0010] S5. Divide the gene modules according to the clustering results; perform principal component analysis on each clustering module to obtain the first principal component, then use the sample label and the first principal component to calculate the correlation coefficient, and finally take the module genes with high correlation as markers of the gene network, which can be used for subsequent downstream analysis.

[0011] According to one embodiment of the present invention, in the step S1, pre-processing includes data cleaning, missing value filling, standardization and variance screening. Preferably, it includes first performing mean screening on the data, then performing variance screening, sorting the variance of the gene, and screening out genes with high variance to ensure the accuracy and comparability of the data. In a preferred embodiment, the homogeneous screening threshold is the fpkm mean of 0.5, and the first 10,000 genes with high variance are screened out.

[0012] According to an embodiment of the present invention, in step S2, genes are used as nodes, the Pearson correlation coefficient C is calculated for all gene pairs, and then the P value is obtained by calculating the correlation coefficient through a given soft threshold st to determine the edge. If the P value is greater than the hard threshold ht setting value, the edge is entered into the gene network diagram, and the P value is used as the weight of the edge. The P value calculation formula is as follows: P = |Cij|^st. In a preferred embodiment, the given soft threshold st = 7, and the hard threshold ht setting value is 0.01.

[0013] According to an embodiment of the present invention, in step S3, the GraphSAGE model includes an input layer, two convolutional layers SAGEConv and an output layer; the input layer accepts the node feature matrix x, each convolutional layer SAGEConv is followed by a network layer for normalizing the features, increasing the model nonlinearity and preventing overfitting, and finally the output layer normalizes the feature dimension to obtain the final graph embedding, which is represented as a low-dimensional node feature. In a preferred embodiment, the number of sampled neighbors of the two convolutional layers is 10, and only 128 nodes are sampled each time, and the output feature dimension after the two convolutional layers SAGEConv is 128; the output layer normalizes the feature dimension to 5.

[0014] According to one embodiment of the present invention, in step S4, the features extracted by the graph convolutional network are clustered using a Gaussian mixture model to divide the genes into different co-expression modules, wherein the number of clusters k is set to twice the square root of the number of graph embeddings, and the covariance type is set to a diagonal type to allow the dimensions of each cluster to be uncorrelated, and the prior weight concentration 1 / k is adjusted according to the maximum value of the number of clusters to control the weight distribution between different clusters. And soft clustering is used to assign a probability distribution belonging to each cluster to each data point.

[0015] According to one embodiment of the present invention, in step S5, the significance score of the clustering module is calculated, principal component analysis is performed on each clustering module to obtain the first principal component, and then the Pearson correlation coefficient is calculated using the sample label and the first principal component. Finally, the module genes with high correlation are taken as markers of the gene network, which can be used for subsequent downstream analysis.

[0016] In another aspect of the present invention, a gene co-expression network identification system based on a graph convolutional neural network is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the above-mentioned gene co-expression network identification method based on a graph convolutional neural network when loaded into the processor.

[0017] Compared with the prior art, the present invention has the following beneficial effects:

[0018] Through the analysis method provided by the present invention, researchers can more effectively extract valuable information from massive amounts of gene expression data, gain a deeper understanding of gene regulatory networks, discover new gene functions and regulatory mechanisms, and provide important guidance for the treatment and prevention of diseases. At the same time, the gene co-expression network identification method of the present invention also has good scalability and flexibility, and can adapt to the analysis needs of gene expression data of different scales and types. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] Figure 1 The figure is a schematic diagram of the technical process of an embodiment of the present invention. DETAILED DESCRIPTION

[0020] In order to facilitate the understanding of the present invention, the present invention will be described more fully below with reference to the relevant drawings. The preferred embodiments of the present invention are given in the drawings. However, the present invention can be implemented in many different forms and is not limited to the embodiments described herein. On the contrary, the purpose of providing these embodiments is to make the understanding of the disclosure of the present invention more thorough and comprehensive.

[0021] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as those commonly understood by those skilled in the art of the present invention. The terms used in the specification of the present invention herein are only for the purpose of describing specific embodiments and are not intended to limit the present invention.

[0022] Unless otherwise specified, the reagents, materials, and equipment used in the examples of the present invention are all commercially available; the experimental methods, unless otherwise specified, are all conventional experimental methods in the art.

[0023] Reference Figure 1 The present invention provides a gene co-expression network identification method based on graph convolutional neural network, comprising the following steps:

[0024] S1. Data preprocessing.

[0025] Preprocessing can include data cleaning, missing value filling, standardization, and variance screening, etc., which can be selected as needed to reduce data noise and scale differences.

[0026] Before performing variance screening, the data is first screened for mean to ensure the accuracy and comparability of the data.

[0027] In one embodiment, the data is firstly screened by mean, with the screening threshold being the fpkm mean of 0.5, and then the variance is screened, the variance of the genes is sorted, and 10,000 genes with high variance are screened out to ensure the accuracy and comparability of the data.

[0028] S2. Construct a gene co-expression network.

[0029] Based on the preprocessed gene expression data, the gene network graph is constructed by calculating the correlation between genes (such as the Pearson correlation coefficient). In the graph, genes are used as nodes and the correlation between genes is used as the weight of the edge.

[0030] Specifically, taking genes as nodes, the Pearson correlation coefficient C is calculated for all gene pairs, and then the P value is obtained by calculating the correlation coefficient through a given soft threshold st to determine the edge. If the P value is greater than the hard threshold ht setting value, the edge is recorded in the gene network diagram, and the P value is used as the weight of the edge. The P value calculation formula is as follows: P = |Cij|^st. In a preferred embodiment, the given soft threshold st = 7, and the hard threshold ht setting value is 0.01.

[0031] S3. Low-dimensional feature expression

[0032] The graph convolutional neural network GraphSAGE model is used to capture node information and extract features from the gene network graph, and the sampler is used to sample and aggregate the neighbors of the nodes. GraphSAGE can learn the features of nodes and their neighboring nodes, thereby obtaining the complex regulatory relationships between genes.

[0033] In one embodiment, the GraphSAGE model includes an input layer, two convolutional layers SAGEConv and an output layer; the input layer accepts a node feature matrix x, each convolutional layer SAGEConv is followed by a network layer for normalizing the features, increasing the model nonlinearity and preventing overfitting, and finally the output layer normalizes the feature dimension to obtain the final graph embedding, which is represented as a low-dimensional node feature. In one embodiment, the number of sampled neighbors of the two convolutional layers is 10, and only 128 nodes are sampled each time. After the two convolutional layers SAGEConv, the output feature dimension is 128; the output layer normalizes the feature dimension to 5.

[0034] S4. Cluster analysis using Gaussian mixture models.

[0035] The Gaussian mixture model (GMM) is used to perform cluster analysis on the features extracted by the graph convolutional network. GMM can fit the complex distribution of data and identify different gene modules. The clustering results divide the genes into different co-expression modules.

[0036] In one embodiment, the number of clusters k is set to twice the square root of the number of graph embeddings, the covariance type is set to diagonal type to allow the dimensions of each cluster to be uncorrelated, and the prior weight concentration 1 / k is adjusted according to the maximum number of clusters to control the weight distribution between different clusters. And soft clustering is used to assign a probability distribution belonging to each cluster to each data point.

[0037] S5. Divide gene modules according to clustering results.

[0038] Principal component analysis is performed on each clustering module to obtain the first principal component, and then the sample label and the first principal component are used to calculate the correlation coefficient, such as the Pearson correlation coefficient; the sample label and the first principal component are then used again, and finally the module genes with high correlation are taken as markers of the gene network, which can be used for subsequent downstream analysis.

[0039] The present invention also provides a gene co-expression network identification system based on a graph convolutional neural network, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the computer program implements the above-mentioned gene co-expression network identification method based on a graph convolutional neural network when loaded into the processor.

[0040] The technical features of the above-described embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above-described embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0041] The above-mentioned embodiments only express several implementation methods of the present invention, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the invention patent. It should be pointed out that, for ordinary technicians in this field, several variations and improvements can be made without departing from the concept of the present invention, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention shall be subject to the attached claims.

Claims

1. A gene co-expression network identification method based on graph convolutional neural network, characterized in that: The steps include: S1. Data preprocessing; S2. Construct a gene network diagram based on gene expression correlation; S3. Capture node information and extract features through the graph convolutional neural network GraphSAGE model; S4. Cluster analysis using Gaussian mixture model; S5. Divide gene modules according to clustering results.

2. The gene co-expression network identification method according to claim 1, characterized in that: In step S1, preprocessing includes data cleaning, missing value filling, standardization and variance screening.

3. The gene co-expression network identification method according to claim 2, characterized in that: The data are first homogenized and then variance screened; preferably, the homogenization screening threshold is an FPKM mean of 0.5, and the variance screens out 10,000 genes with high variance.

4. The gene co-expression network identification method according to claim 1, characterized in that: In step S2, genes are used as nodes, and the Pearson correlation coefficient C is calculated for all gene pairs. The correlation coefficient is then calculated using a given soft threshold st to obtain the P value to determine the edge. If the P value is greater than the hard threshold ht setting value, the edge is entered into the gene network diagram, and the P value is used as the weight of the edge. The P value calculation formula is as follows: P = |Cij|^st.

5. The gene co-expression network identification method according to claim 4, characterized in that: The given soft threshold st=7, and the hard threshold ht setting value is 0.

01.

6. The gene co-expression network identification method according to claim 1, characterized in that: In step S3, the GraphSAGE model includes an input layer, two convolutional layers SAGEConv and an output layer; the input layer accepts the node feature matrix x, and each convolutional layer SAGEConv is followed by a network layer for normalizing the features, increasing the model nonlinearity and preventing overfitting. Finally, the output layer normalizes the feature dimensions to obtain the final graph embedding, which is represented as low-dimensional node features.

7. The gene co-expression network identification method according to claim 6, characterized in that: The number of sampled neighbors in both convolutional layers is 10, and only 128 nodes are sampled each time. The output feature dimension after two convolutional layers SAGEConv is 128; the output layer normalizes the feature dimension to 5.

8. The gene co-expression network identification method according to claim 1, characterized in that: In step S4, the number of clusters k is set to twice the square root of the number of graph embeddings, the covariance type is set to diagonal, the prior weight concentration 1 / k is adjusted according to the maximum number of clusters, and soft clustering is used to assign a probability distribution belonging to each cluster to each data point.

9. The gene co-expression network identification method according to claim 1, characterized in that: In step S5, principal component analysis is performed on each clustering module to obtain the first principal component, and then the Pearson correlation coefficient is calculated using the sample label and the first principal component. Finally, the module genes with high correlation are taken as markers of the gene network for subsequent downstream analysis.

10. A gene co-expression network identification system based on graph convolutional neural network, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the computer program is loaded into a processor, the method for identifying gene co-expression networks based on a graph convolutional neural network according to any one of claims 1 to 9 is implemented.

Images