Employing functional interactions for characterization and detection of sparse complexes from yeast PPI networks

TL;DR

This paper introduces a new method that uses functional interactions to improve the detection of sparse protein complexes in yeast PPI networks, significantly increasing the recovery rate of known complexes.

Contribution

It proposes the CE score to measure complex derivability and an algorithm SPARC that enhances detection of sparse complexes by leveraging functional interactions.

Findings

Recovered 104 out of 123 known complexes, a 47% improvement.

Effectively distinguishes sparse from dense complexes using CE score.

Enhances existing methods to detect previously missed complexes.

Abstract

Over the last few years, several computational techniques have been devised to recover protein complexes from the protein interaction (PPI) networks of organisms. These techniques model "dense" subnetworks within PPI networks as complexes. However, our comprehensive evaluations revealed that these techniques fail to reconstruct many 'gold standard' complexes that are "sparse" in the networks (only 71 recovered out of 123 known yeast complexes embedded in a network of 9704 interactions among 1622 proteins). In this work, we propose a novel index called Component-Edge (CE) score to quantitatively measure the notion of "complex derivability" from PPI networks. Using this index, we theoretically categorize complexes as "sparse" or "dense" with respect to a given network. We then devise an algorithm SPARC that selectively employs functional interactions to improve the CE scores of predicted complexes, and thereby elevates many of the "sparse" complexes to "dense". This empowers existing methods to detect these "sparse" complexes. We demonstrate that our approach is effective in reconstructing significantly many complexes missed previously (104 recovered out of the 123 known complexes or ~47% improvement).