CUBCO+: prediction of protein complexes based on min-cut network partitioning into biclique spanned subgraphs

Abstract

High-throughput proteomics approaches have resulted in large-scale protein–protein interaction (PPI) networks that have been employed for the prediction of protein complexes. However, PPI networks contain false-positive as well as false-negative PPIs that affect the protein complex prediction algorithms. To address this issue, here we propose an algorithm called CUBCO+ that: (1) employs GO semantic similarity to retain only biologically relevant interactions with a high similarity score, (2) based on link prediction approaches, scores the false-negative edges, and (3) incorporates the resulting scores to predict protein complexes. Through comprehensive analyses with PPIs from Escherichia coli, Saccharomyces cerevisiae, and Homo sapiens, we show that CUBCO+ performs as well as the approaches that predict protein complexes based on recently introduced graph partitions into biclique spanned subgraphs and outperforms the other state-of-the-art approaches. Moreover, we illustrate that in combination with GO semantic similarity, CUBCO+ enables us to predict more accurate protein complexes in 36% of the cases in comparison to CUBCO as its predecessor.

An Evolutionary Algorithm With Heuristic Operator for Detecting Protein Complexes in Protein Interaction Networks With Negative Controls

“…Moreover, interactions with low confidence values may be discarded in subsequent analyses. However, different topological metrics and link prediction algorithms [12,27] can be used to score false-negative interactions and contribute the top-scored ones to PPI networks [38]. There is a general agreement among biologists that proteins that are closely located to one another in the PPI network are perform similar functions, and genes that are regulated by the same transcription factors likely to have activities that are substantially similar to one another (genes causing similar diseases).…”