Identifying dense subgraphs in protein–protein interaction network for gene selection from microarray data

Abstract

Selection of important genes responsible for a disease is an important task in bioinformatics. Microarray data are often used with differential expression being considered as a cue. Recently, such expression data are supplemented by gene ontology and genes/proteins interaction network for the selection task. The functional knowledge and interaction structure have become critical for understanding the biological processes, including selection of genes potentially associated to complex diseases. In this paper, we propose an approach that combines expression analysis with structural analysis of protein–protein interaction networks to identify genes associated with complex diseases. The dense subgraph structures embedded in the networks are extracted. We present results on three different types of benchmark cancer dataset (prostate cancer, interstitial lung disease and chronic lymphocytic leukemia) and show that several interesting biological information could be inferred, besides achieving a high prediction accuracy. The proposed methodology helps to identify not just differentially expressed genes but also hub genes important in biological processes.