Prioritization of Nonsynonymous Single Nucleotide Variants for Exome Sequencing Studies Via Integrative Learning on Multiple Genomic Data

Abstract

The rapid advancement of next generation sequencing technology has greatly accelerated the progress for understanding human inherited diseases via such innovations as exome sequencing. Nevertheless, the identification of causative variants from sequencing data remains a great challenge. Traditional statistical genetics approaches such as linkage analysis and association studies have limited power in analyzing exome sequencing data, while relying on simply filtration strategies and predicted functional implications of mutations to pinpoint pathogenic variants are prone to produce false positives. To overcome these limitations, we herein propose a supervised learning approach, termed snvForest, to prioritize candidate nonsynonymous single nucleotide variants for a specific type of disease by integrating 11 functional scores at the variant level and 8 association scores at the gene level. We conduct a series of large-scale in silico validation experiments, demonstrating the effectiveness of snvForest across 2,511 diseases of different inheritance styles and the superiority of our approach over two state-of-the-art methods. We further apply snvForest to three real exome sequencing data sets of epileptic encephalophathies and intellectual disability to show the ability of our approach to identify causative de novo mutations for these complex diseases. The online service and standalone software of snvForest are found at http://bioinfo.au.tsinghua.edu.cn/jianglab/snvforest.