Ranking Docking Models by Coevolution Analysis

Abstract

Protein-protein interactions are critical for diverse cell processes. One of the most used approaches to solve these protein-protein interactions at the atomic level is protein docking. The generation of false positive hits is however well documented drawback of docking algorithms, due the limitations of the scoring function in describing solvation energies and protein flexibility. Protein coevolution analysis can be useful to contour this problem. Protein coevolution arises from the concept of macromolecular coevolution and refers to amino acids correlated mutations that stems from natural selection. The trail of coevolution can be find in a multiple sequence alignment (MSA), by analysis which measure the entropy of its columns or the correlations of the species homologous sequences. Thus, we combined these analyses and docking calculations to improve docking modeling. Looking for new developments in the field, we have created a software, the Docking Score Module (DSM). Specifically, DSM takes as input 3D models generated from docking programs and a based on proteins models. At the heart of the algorithm, new sub-alignments exclusively containing amino acid positions which are in protein-protein interface are generated for each docking model. For each new sub-alignment, DSM calculates mutual information (MI) as defined from Shannon's information entropies, and correlation index (r), which reports phylogenetic similarity of protein sequences in MSA. The output consists of a text file with job information and a plot, in which models are weighted by MI or r. For the case studies, we docked eleven complexes using different docking web servers. Across a space of docking models and taking into consideration the value of MI and r, best ranked models in DSM contain the biggest number of native contacts and are closer to the target complex.