Comparison of tissue/disease specific integrated networks using directed graphlet signatures

Analysis of integrated genome-scale networks is a challenging problem due to heterogeneity of high-throughput data. There are several topological measures, such as graphlet counts, for characterization of biological networks.

In this paper, we present methods for counting small sub-graph patterns in integrated genome-scale networks which are modeled as labeled multidigraphs. We have obtained physical, regulatory, and metabolic interactions between H. sapiens proteins from the Pathway Commons database. The integrated network is filtered for tissue/disease specific proteins by using a large-scale human transcriptional profiling study, resulting in several tissue and disease specific sub-networks. We have applied and extended the idea of graphlet counting in undirected protein-protein interaction (PPI) networks to directed multi-labeled networks and represented each network as a vector of graphlet counts. Graphlet counts are assessed for statistical significance by comparison against a set of randomized networks. We present our results on analysis of differential graphlets between different conditions and on the utility of graphlet count vectors for clustering multiple condition specific networks.

Our results show that there are numerous statistically significant graphlets in integrated biological networks and the graphlet signature vector can be used as an effective representation of a multi-labeled network for clustering and systems level analysis of tissue/disease specific networks.