Research On The Methods Of Gene Regulatory Network Construction

With the rapid development of the next generation gene sequencing technology,the amount of gene expression data has gradually increased in recent years.Gene expression data contain abundant information of gene activities.It is very important to understand the life mechanism in biological cells by mining the regulatory relationships among genes from gene expression data.Constructing the gene regulatory networks can contribute to reveal the regulatory relationships among genes and provide a visual tool for the analysis of gene expression data.Static gene expression data reflect the expression levels of genes in a stable state,which can indicate the current physiological state of cells.Static gene expression data have the characteristics of high dimension and small samples.It is a challenging task to accurately detect the regulatory relationships among genes in the case of high dimension and small samples.In this thesis,DSWLasso algorithm is proposed to construct the gene regulatory networks by using iterative weighted lasso.DSWLasso uses the conception of distance correlation to generate distance samples,which increases the number of samples.Then it uses multivariate linear regression to construct the initial gene regulatory networks,and improves the accuracy of the constructed gene regulatory networks by using iterative weighted lasso to delete the weak regulatory relationships.The experimental results on the simulated Gaussian data and 25 gene expression data sets showed that the accuracy of the gene regulatory networks constructed by DSWLasso is superior to those by ARACNE,GeneNet,DPM,reg-DPM and PMI in most cases.The life activities in biological cells are dynamics,and the regulatory effects among genes are not always simultaneous,but have a certain time delay.Time-series gene expression data contain the information of gene expression level changing over time.Therefore,mining the regulatory relationships among genes from time-series gene expression data can explain the activities of genes in different states comprehensively and accurately.This thesis proposes GAELM algorithm,which combines genetic algorithm and extreme learning machine to identify the gene regulation relationships from time-series gene expression data.Genetic algorithm can search the regulatory genes of a target gene.Extreme learning machine has the advantage of fast training speed and can be used to obtain the regulation relationships among genes.In order to improve the accuracy of the regulation relationships detected by extreme learning machine,GAELM uses dynamic time warping to preliminarily select potential target genes.On real time-series gene expression data sets and simulated time-series data sets,the gene regulatory networks constructed by GAELM have a better performance than those by ELM-GRNNminer,HRNN,TD-ARACNE,BANJO and GRNNminer in most cases.DSWLasso and GAELM algorithm can detect the regulatory relationships among genes and construct gene regulatory networks from static and time-series gene expression data respectively.The experimental results on the corresponding real gene expression data sets and the simulated data sets showed the validation of these two methods.