Exploiting Co-evolution across Protein Families for Predicting Native Contacts and Protein-Protein Interaction Surfaces

Abstract

Correlated substitution patterns between residues of a protein family have been exploited to reveal information on the structures of proteins. However, such studies require a large number (e.g., the order of one thousand) of homologous yet variable protein sequences. So far, most studies have been limited to a few exemplary proteins for which a large number of such sequences happen to be available. Rapid advances in genome sequencing will soon be able to generate this many sequences for the majority of common bacterial proteins. Sequencing a large number of simple eukaryote such as yeast can in principle generate similar number of common eukaryotic protein sequences, beyond a collection of highly amplified protein domains which already reach the necessary numbers.

The heart of our approach is a novel, statistical-physics inspired analysis of residue co-evolution, which has recently been shown (i) to accurately predict inter- and intra-protein amino-acid spatial dependencies, (ii) to achieve structural predictions with experimental accuracy when integrated with molecular simulations.

A systematic evaluation of the information contained in correlated substitution patterns for predicting residue contacts will be given, as a first step towards a purely sequence-based approach to protein structure, and protein-protein interaction prediction.