Predicting residue and helix contacts in membrane proteins

Abstract

Helix-helix contacts are an important feature of alpha-helical membrane proteins as they defi ne their characteristic helix bundle structure. No bioinformatics approaches for the prediction of pairwise residue contacts in membrane proteins have existed until recently. In this chapter we describe novel contact prediction methods based on residue coevolution and machine learning techniques specifi cally geared towards membrane proteins. While contact prediction accuracies are limited to ~10% using co-evolving residues alone, machine learning methods are able to improve these accuracies signifi cantly to more than 25% by using available membrane protein structures as a training dataset and incorporating membrane protein specifi c sequence features into the prediction process. Importantly, predicted residue contacts allow for identifi cation of interacting transmembrane helices with high accuracy. As diff erent membrane protein structures can be distinguished by their specifi c patt ern of helix interactions, predicted residue contacts may not only serve as structural constraints in modeling experiments, but also constitute valuable information for structural classifi cation of membrane proteins with unknown structure.