Prediction Of Cancer Driver Genes Based On Gene Expression Proifles And Molecular Interaction Networks

Topic in this paper comes from the Ministry of Health deployed (management)hospital clinical subjects focus projects “Research and Application of Prognostic RiskAssessment and Early Warning in the Cancer”. We select the prediction of cancerdriver genes as our research direction by talking with the cooperators and extensivelyresearching the main problems and development in this field at home and abroad.The morbidity and mortality of cancer increased year by year worldwide, whichhas a serious negative impact on the patients themselves, their families, and even thewhole society. In the process of tumor formation, mutations of cancer driver genesplay central roles in the genetic mechanism of cancer development. So the predictionof cancer driver genes has the important academic value and the widespreadapplication value in cancer clinical treatment. However, this work are still challengingby the complex mechanics of cancer and the limitations of avaliable computationaland experimental methods.In this paper, we present a novel system biology model, based on the informationfrom gene expression profiles and molecular interaction networks, to predict cancerdriver genes carried by specific tumor samples. Our model first use the paried t-testsand fold-change method to identify the differentially expressed cancer genes. Thenconstruct a cancer related sub-network according to the molecular biological conceptsand the local and central hypotheses about molecular interaction networks. Next,calculate the Pearson’s correlation coefficients for all the gene-gene interactions in thesub-network, and then use the Wilcoxon’s signed rank tests to select the genes whosecorrelation coefficients were significant changes between tumor and normalconditions. The model finally output a set of predicted genes that highly correlated with the specific phenotypes of cancer samples.We applied the model to predict driver genes in breast and colorectal cancers, andgot28and24predicted cancer driver genes respectively. By comparison with thepublished experiment results, we found about93%and83%genes in the predictionsets have already known to be related to the corresponding cancer types, and some ofthem are widely recognized cancer driver genes. We further did the enrichmentanalysis on the predition sets, results show these genes are significantly enriched inthe GO biological processes and KEGG pathways related to cancer. Moreover, bycomparison with the widely accepted gene list about the cancer driver genes, wefound the prediction accuracy of our model is much better than the method withoutintegrate molecular interaction networks.In conclusion, our work proceeds from the identification of differentiallyexpressed genes in cancer, but go further to predict the cancer driver genes byintegrating informations from molecular interaction networks. This effort can offersome valuable guidances to the next step studies like the design of expreimentalvalidations and the construction of quantitative cancer models in this project, and willeventually help to improve the therapeutic and treatment stategies for cancer patientsin the future.