Data Analysis And Network Construction Of Complex Diseases

Because complex diseases involve multiple genes,environments and some other unknown random factors,it is very difficult to understand and know these diseases.With the progress of science and technology and the gradual maturity of highthroughput sequencing technology,a large number of data from different omics have emerged.How to make full use of these big data from different omics for integrated analysis,in order to understand the pathogenesis of diseases and predict the prognosis of patients,has become a challenging problem.Based on the gene expression data and imaging data of children’s respiratory diseases and gliomas,we make the integrative analysis through combined with network construction and machine learning algorithm,the main innovative work in this thesis is described as follows:1.Based on the gene expression data of respiratory syncytial virus(RSV)infection in children with respiratory diseases,a complex consisting of three genes USP18,ISG15 and IFIT2 is extracted as a biomarker of immune enhancement diseases by constructing a dynamic gene regulatory network and identifying high impact modules.2.Through the integration of genomic data and imaging data of patients with low glioma,a new quantitative index is designed to classify the low-risk and high-risk patients’ subgroups of cancer.Firstly,the network construction of gene expression data and the feature extraction of image data by Cox regression model are obtained.12 network modules are obtained and 8 imaging biomarkers are identified.Then the correlation analysis is carried out.Finally,a new prognosis prediction index is proposed.The optimal weighting coefficient is calculated by particle swarm optimization algorithm,and the consistency index(c-index)is predicted by fusion index The results show that the prediction accuracy is 72.1%and 0.8071,which is better than that of the single group data before integration.The concordance index of the final integrated index is 0.8071 and the accuracy of prediction arrive at 72.1%,both higher than any single feature from genomic data or radiomics data.