Kevlar: a mapping-free framework for accurate discovery ofde novovariants
AbstractMotivationDiscovery of genetic variants by whole genome sequencing has proven a powerful approach to study the etiology of complex genetic disorders. Elucidation of all variants is a necessary step in identifying causative variants and disease genes. In particular, there is an increased interest in detection ofde novovariation and investigation of its role in various disorders. State-of-the-art methods for variant discovery rely on mapping reads from each individual to a reference genome and predicting variants from difference observed between the mapped reads and the reference genome. This process typically results in millions of variant predictions, most of which are inherited and irrelevant to the phenotype of interest. To distinguish between inherited variation and novel variation resulting fromde novogermline mutation, whole-genome sequencing of close relatives (especially parents and siblings) is commonly used. However, standard mapping-based approaches tend to have a high false-discovery rate forde novovariant prediction, which in many cases arises from problems with read mapping. This is a particular challenge in predictingde novoindels and structural variants.ResultsWe have developed a mapping-free method, Kevlar, forde novovariant discovery based on direct comparison of sequence content between related individuals. Kevlar identifies high-abundancek-mers unique to the individual of interest and retrieves the reads containing thesek-mers. These reads are easily partitioned into disjoint sets by sharedk-mer content for subsequent locus-by-locus processing and variant calling. Kevlar also utilizes a novel probabilistic approach to score and rank the variant predictions to identify the most likelyde novovariants. We evaluated Kevlar on simulated and real pedigrees, and demonstrate its ability to detect bothde novoSNVs and indels with high sensitivity and specificity.Availabilityhttps://github.com/kevlar-dev/kevlar