Item

Abstract

High-throughput approaches have generated large-scale protein-protein interaction (PPI) networks that are used in prediction of protein complexes. Here, we introduce CUBCO—a minimum cut-based algorithm that predicts protein complexes as biclique spanned subgraphs while relying on link prediction approaches to score and incorporate missing interactions. Our comprehensive analyses with PPIs from different organisms show that CUBCO performs on par with the best-performing approaches, that model protein complexes as biclique spanned subgraphs, and outperforms the remaining contenders. We also show that the usage of link prediction approaches in CUBCO improves the prediction of protein complexes on average 34.22% in all comparisons. Finally, CUBCO recovers ~40% and ~11% of known protein complexes from the Pan-Plant and Metazoan PPI networks. Therefore, CUBCO represents an efficient, parameter-free approach for accurate prediction of protein complexes from PPI networks.