Well-connectedness - a novel measure for improving protein complex detection from PPI network

Recently, protein complex prediction via computational approaches from PPI network data has gained significant popularity among bioinformatics researchers. Majority of these approaches are unsupervised. The quality measures they use to guide the mining process have some drawbacks in reflecting the properties of a protein complex presented in the PPI network data. In this paper, we propose a novel measure, named well-connectedness. It more accurately measures the similarity of interactions between a mined protein complex and a real protein complex. We have incorporated well-connectedness measure in two state-of-the-art overlapping protein complex prediction algorithms. Empirical analysis on four real world data sets has shown that, considering our measure enables to mine more meaningful protein complexes.