Meta-DPI: A Computational Metamethod for Predicting Protein-Protein Interfaces.

Abstract

Protein-protein interactions (PPIs) regulate many biological processes that are integral for the survival, function, growth, and evolution of organisms. Determination of the amino acid residues at these interaction sites, or interfaces, enhances our understanding of the molecular mechanisms by which proteins carry out their functions, and it facilitates the development of therapeutics by identifying critical sites for disrupting protein function. Due to the timeconsuming and costly nature of experimental approaches for identifying interacting residues, computational methods are employed to streamline the process. This work describes the development of meta-DPI, a computational metamethod for predicting protein-protein interfaces that integrates the orthogonal prediction methods DockPred, PredUs 2.0, and ISPRED4. Crossvalidation experiments on two datasets containing a total of 223 protein complexes illustrate that meta-DPI significantly outperforms each of its three constituent methods (DockPred, PredUs 2.0, and ISPRED4) as measured by both single-threshold and threshold-free evaluation metrics. The enhanced predictive power of meta-DPI demonstrates how metamethods create improved interface predictors, which in turn, enable molecular mechanisms to be understood more thoroughly and drug targets to be identified more efficiently.