Improved network-based identification of protein orthologs.

Motivation: Identifying protein orthologs is an important task that is receiving growing attention in the bioinformatics literature. Orthology detection provides a fundamental tool towards understanding protein evolution, predicting protein functions and interactions, aligning protein-protein interaction (PPI) networks of different species and detecting conserved modules within these networks.

Results: Here, we present a novel diffusion-based framework that builds on the Rankprop algorithm for protein orthology detection and enhances it in several important ways. Specifically, we enhance the Rankprop algorithm to account for the presence of multiple paralogs, utilize PPI, and consider multiple (>2) species in parallel. We comprehensively benchmarked our framework using a variety of training datasets and experimental settings. The results, based on the yeast, fly and human proteomes, show that the novel enhancements of Rankprop provide substantial improvements over its original formulation as well as over a number of state of the art methods for network-based orthology detection.

Availability: datasets and source code are available upon request.