Identification Of Phenotype Associated Molecular Biological Subnetworks Based On Optimization Method

Among various phenotypes, disease is one of the most important to human beings. Asthere are complex functional relationships among distinct molecules within cells, thedysfunction of one single gene will not only influences its product, but also affects othergenes in the biological systems. Therefore, it is crucial to utilize molecular interactionnetworks to identify the network markers associated with phenotypes from the systematicperspective. In this thesis, I present several works to identify phenotype associatedmolecular pathways and network modules, which can be found below.Firstly, we summarized the computational approaches for identifying phenotypeassociated molecular pathways based on molecular interaction networks. In particular, thoseapproaches were categorized according to the following information: the source of data,whether the background network is known, whether there are feedback loops in theidentified subnetwork and whether the resultant subnetwork is oriented. As a result, thesecomputational approaches were grouped into three categories depending on themethodologies they exploited, i.e. graph model, optimization model and heuristic method.Secondly, we proposed a network flow model to identify the molecular pathwaysassociated with breast cancer. Based on the gene expression data of breast cancer andhuman protein-protein interaction network, we constructed an integrated molecularinteraction network. Then, we applied our network flow model to identify molecularpathways associated with breast cancer. At last, we proved that the predicted molecularpathways associated with breast cancer are reasonable by functional enrichment analysis ofgenes within the pathways.Lastly, we presented a new mathematical model to identify molecular subnetworks asdrug targets. Based on protein post-translational modification network, we recognized somenetwork motifs which are enriched with drug target proteins. After further analysis, wefound that our identified network motifs have closed relationship with disease. Inspired bythese findings, we proposed a novel method to predict new drug target proteins based onnetwork motifs and achieved good results on the standard data set.