Identification of active disease-associated subnetworks in human protein-protein interaction networks using the MCL algorithm

Official URL: http://risc01.sabanciuniv.edu/record=b1615018 (Table of Contents)

An active subnetwork is a group of highly interacting genes that are associated with a particular disease in a biological interaction network. Finding these subnetworks facilitates the understanding of the molecular mechanisms of diseases and contributes to the process of devising treatment strategies, making the identification of active subnetworks an important problem. In this thesis, the use of a clustering algorithm is proposed for the detection of active subnetworks and a methodology that is based on the Markov Cluster (MCL) algorithm is implemented. The methodology uses graph representation to represent the human protein-protein interaction network, a novel scoring scheme to appoint weights to the interactions among the network, the Markov Cluster algorithm for the active subnetwork search, a scoring formula to assign scores to each found subnetwork and an elimination of subnetworks depending on those scores, followed by a functional enrichment step to discover the functionally important KEGG pathways related with found subnetworks. This methodology is applied on WTCCC Rheumatoid Arthritis (RA) dataset and identified: KEGG pathways previously found to be RA-related (e.g., NF-kappaB, Jak-STAT, Toll-like receptor, MAPK signaling pathways), and additional pathways (e.g., Serotonergic synapse) as associated with RA. The comparative study shows that the presented method outperforms state-of-the-art techniques, and functional enrichment results demonstrate that the method can successfully detect significant subnetworks that are related with RA which is a complex multifactorial disease. Therefore, it is proposed that the method can be used on the datasets of other complex diseases to identify active disease-associated subnetworks.