Prediction and analysis of effects of non-synonymous single-nucleotide polymorphisms on protein function.

摘要

Genetic variation is the raw material of evolution. By selecting the genotypes most "fit" to the current environment, nature determines further biological landscapes of the planet. Organisms of a single species are generally very similar genetically. However, the slight intra-species genome diversity accounts for most of the observed disparities in individual fitness (e.g. disease resistance or tolerance to environmental insults). In all species, a very large fraction of this variation is due to single nucleotide polymorphisms (SNPs).; More precisely defined, SNPs are single position (base) differences between the DNA sequences of two organisms of the same species. They can be generally classified into groups based on their location (coding/non-coding DNA) and effect in protein (synonymous/non-synonymous). Of these, the non-synonymous set is the most biophysically apparent - nsSNPs, by definition, alter the product protein sequence. However, since even these mutants do not necessarily have functional/structural consequences, the evaluation of the extent of their influence is non-trivial.; NsSNPs are responsible for numerous diseases. For example, a point mutation in the hemoglobin beta gene is one proven cause of sickle cell anemia. Other diseases, such as diabetes and cancer, have been correlated with a number of SNPs but their true genetic mechanisms remain unclear. "Wet-lab" experiments designed to evaluate the functional consequences of mutations are time-consuming and may be costly. For these reasons, in silico methods have been developed that try to separate the non-neutral mutations (having an effect on protein function) from neutral ones (no effect). Our contribution to this field is a neural-network based method called SNAP (Screening for Non-Acceptable Polymorphisms). This tool classifies nsSNPs into non-neutral/neutral categories using only the information extracted from corresponding protein sequences. SNAP achieves high levels of accuracy by combining conservation/family data with sequence-based predictions of various protein features, such as secondary structure and solvent accessibility.; This work describes our method in detail, outlines the assessment of performance using various data sets, and addresses the relationship of SNAP predictions to diseases, protein structure, and functional site annotations. Information presented here suggests that protein sequence carries enough information for making accurate assumptions regarding protein structure and function.