Research On Identifying Key Glioma Genes Based On Gene Expression Data And Network Models

Glioma accounts for 80%of all malignant brain tumors,which is a serious hazard to human health.However,the current-inadequate acknowledge to the pathogenesis of glioma leads to poor efficacy in the diagnosis and treatment of glioma.A large amount of data with hidden biological information has been generated since the Human Genome Project began.And it is thus a crucial problem that how to identify the key genes related to glioma from this huge data.This paper is based on gene expression data and network models,using data analysis,network topology information mining and network dynamic analysis to identify glioma-related and significant genes,which aims to provide further research for the pathogenesis of glioma.The specific work of this paper is as follows:1)Based on the gene expression data of glioma patients,we apply BMR-ReliefF algorithm combined with SVM-RFE algorithm to filter the characteristic genes,then the5-fold cross-validation method is used to verify its validness?the AUC value is above0.85?,and 25 characteristic genes are identified from it.Go analysis of these 25characteristic genes reveals that the 9 major enriched GO terms have correlations with glioma.And CDK6,NDRG1,MDK genes have been found linked with glioma closely in PubMed literature.From the above,it illuminates that the way we applied in identifying glioma genes is effective.What's more,KEGG pathway analysis suggests that the neuroactive ligand-receptor interactions and related genes could be the crucial biologic-pathways and potentially important genes associated with glioma,which provides the theoretical basis for further exploration on the pathogenesis of glioma.2)As for the disease-gene heterogeneous network,we come up with adopting DeepWalk method to deeply mine the topology information of the network.Then we put forward two strategies to predict the link between glioma and genes,which are DBSI?disease-based similarity inference?and GBSI?gene-based similarity inference?respectively.The result that DBSI is better than GBSI for prediction is proved by 5-fold cross-validation method(AUC_DBSI=0.87,AUC_GBSI=0.81).Accordingly,we identify 30genes that are bound up with glioma very likely by DBSI.On the one hand,biofunctional enrichment analysis indicates that IL1A and HLA-DQA etc.are enriched in cytokine-cytokine receptor interaction signaling pathway and TNF signaling pathway,which offers a theoretical reference for through research on glioma treatment as the potential target.On the other hand,the main enriched biological processes of GO analysis?12?and the KEGG pathways?8?have been confirmed to be strongly related to the pathogenesis of glioma,which verifies the validity of the method used in this chapter.3)Towards the brain-specific gene regulatory network,we study the network dynamics by complex network controllability analysis,and classify the genes according to the change in the number of minimum driving nodes after deleting network nodes.Then the network nodes are divided into two parts,which are critical and ordinary genes separately.Finally,three aspects are analyzed,which are the network topology analysis,the enrichment analysis in biological function database and the robustness analysis of network,showing that the method applied here is robust and the critical genes we find have association with glioma indeed.