Towards detecting protein complexes from protein interaction data

Abstract

High-throughput methods for detecting protein-protein in-teractions (PPI) have given researchers an initial global picture of protein interactions on a genomic scale. These interactions connect proteins into a large protein interaction network (PIN). However, both the size of the data sets and the noise in the data pose big challenges in effectively analyzing the data. In this paper, we investigate the problem of protein complex detection, i.e., finding biologically meaningful subsets of proteins, from the noisy protein interaction data. We identify the difficulties and propose a "seed-refine" approach, including a novel subgraph quality measure, an appropriate heuristics for finding good seeds and a novel subgraph refinement method. Our method considers the properties of protein complexes and the noisy interaction data. Experiments show the effectiveness of our method. © Springer-Verlag Berlin Heidelberg 2006.