Predicting non-neutral missense mutations and their biochemical consequences using genome-scale homology modeling of human protein complexes

TL;DR

This paper introduces a structure-based homology modeling approach to identify disease-causing missense mutations in human proteins, providing biochemical insights and improving prediction accuracy over existing methods.

Contribution

It presents a novel large-scale, structure-based method for detecting non-neutral mutations and their biochemical effects, surpassing sequence conservation approaches.

Findings

Disease mutations are enriched in binding sites

More binding sites identified than in RefSeq

Machine learning predictor outperforms existing methods

Abstract

Computational methods are needed to differentiate the small fraction of missense mutations that contribute to disease by disrupting protein function from neutral variants. We describe several complementary methods using large-scale homology modeling of human protein complexes to detect non-neutral mutations. Importantly, unlike sequence conservation-based methods, this structure-based approach provides experimentally testable biochemical mechanisms for mutations in disease. Specifically, we infer metal ion, small molecule, protein-protein, and nucleic acid binding sites by homology and find that disease-associated missense mutations are more prevalent in each class of binding site than are neutral mutations. Importantly, our approach identifies considerably more binding sites than those annotated in the RefSeq database. Furthermore, an analysis of metal ion and protein-protein binding sites predicted by machine learning shows a similar preponderance of disease-associated mutations in these sites. We also derive a statistical score for predicting how mutations affect metal ion binding and find many dbSNP mutations that likely disrupt ion binding but were not previously considered deleterious. We also cluster mutations in the protein structure to discover putative functional regions. Finally, we develop a machine learning predictor for detecting disease-associated missense mutations and show that it outperforms two other prediction methods on an independent test set.