Evaluation of Structural and Evolutionary Contributions to Deleterious Mutation Prediction

Abstract

Methods for automated prediction of deleterious protein mutations have utilized both structural and evolutionary information but the relative contribution of these two factors remains unclear. To address this, we have used a variety of structural and evolutionary features to create simple deleterious mutation models that have been tested on both experimental mutagenesis and human allele data. We find that the most accurate predictions are obtained using a solvent-accessibility term, the C^β density, and a score derived from homologous sequences, SIFT. A classification tree using these two features has a cross-validated prediction error of 20.5% on an experimental mutagenesis test set when the prior probability for deleterious and neutral cases is equal, whereas this prediction error is 28.8% and 22.2% using either the C^β density or SIFT alone. The improvement imparted by structure increases when fewer homologs are available: when restricted to three homologs the prediction error improves from 26.9% using SIFT alone to 22.4% using SIFT and the C^β density, or 24.8% using SIFT and a noisy C^β density term approximating the inaccuracy of ab initio structures modeled by the Rosetta method. We conclude that methods for deleterious mutation prediction should include structural information when fewer than five to ten homologs are available, and that ab initio predicted structures may soon be useful in such cases when high-resolution structures are unavailable.

Introduction

The rapid discovery of single nucleotide polymorphisms (SNP) in the human genome has created an opportunity for high-throughput deleterious mutation prediction methods to discover and prioritize candidate human disease alleles from the pool of uncharacterized non-synonymous coding SNPs. Several methods have recently been developed specifically for this purpose,1., 2., 3. which utilize information from both evolution and protein structure. In principle, a protein mutation could be deleterious either because it destabilizes structure or it disrupts a functional site involved in catalysis, ligand-binding, or interaction with another protein. For this reason, one might expect evolutionary and structural information from proteins to complement one another: structural information should help to identify destabilizing mutations, while highly conserved positions in multiple sequence alignments (msa) can help to identify functional sites.⁴ While this principle is qualitatively clear, the relative importance and complementarity of these feature domains has not been well characterized with regard to computational prediction of intolerant mutations. Such characterization is important both for improving the accuracy of predictive models and for determining the relative error of predictions made when sequence or structural information is reduced or absent, as is often the case when few sequence homologs exist or only a coarse approximation to the protein structure is available.

In this study, we investigate the relative strength of a variety of structural and evolutionary features described in previous work and examine how the most effective features complement one another in simple classification models. We characterize the performance of classifiers as a function of the number of homologs available for the calculation of evolutionary features, which indicates how the reliability of deleterious mutation prediction is affected by the common problem of having few homologous sequences available. Finally, we explore the possibility of improving classification for cases where no experimental structure is available, by using structures generated from ab initio prediction methods.