Reducing the search space for causal genetic variants with VASP

Abstract

Motivation: Increasingly, cost-effective high-throughput DNA sequencing technologies are being utilized to sequence human pedigrees to elucidate the genetic cause of a wide variety of human diseases. While numerous tools exist for variant prioritization within a single genome, the ability to concurrently analyze variants within pedigrees remains a challenge, especially should there be no prior indication of the underlying genetic cause of the disease. Here, we present a tool, variant analysis of sequenced pedigrees (VASP), a flexible data integration environment capable of producing a summary of pedigree variation, providing relevant information such as compound heterozygosity, genome phasing and disease inheritance patterns. Designed to aggregate data across a sequenced pedigree, VASP allows both powerful filtering and custom prioritization of both single nucleotide variants (SNVs) and small indels. Hence, clinical and research users with prior knowledge of a disease are able to dramatically reduce the variant search space based on a wide variety of custom prioritization criteria.

Availability and implementation: Source code available for academic non-commercial research purposes at https://github.com/mattmattmattmatt/VASP.

Contact: matt.field@anu.edu.au

Supplementary information: Supplementary data are available at Bioinformatics online.

1 Introduction

While exome sequencing has been successfully utilized in the discovery of causal variation using small numbers of unrelated individuals (Ng et al., 2010) it is becoming clear that this approach is insufficient to reliably identify the genetic causes in many cases (Yang et al., 2013). Increasingly, genetic variation information from related individuals is being employed to reduce the search space for causal variants, by both the prioritization of variants common to affected individuals and the exclusion of benign variants shared between affected and unaffected individuals. The effective analysis of sequenced pedigrees requires new tools capable of combining variant specific and pedigree wide annotations with powerful filtering options to dramatically reduce the causal variation search space. Current tools focus on either progressively removing variants based on criteria deemed unlikely to be causal (Li et al., 2012) or by focusing on variants matching specific inheritance models [compound heterozygotes (Kamphans et al., 2013); autosomal dominant (Koboldt et al., 2014)]. Here, we present variant analysis of sequenced pedigrees (VASP), a tool that integrates and summarizes information across a sequenced pedigree without making any assumptions regarding disease inheritance further providing the full integrated pedigree variation information for subsequent prioritization. VASP is standalone software written in Perl derived from our original variant detection pipeline (Andrews et al., 2012).

2 Methods

VASP integrates variant information from each pedigree member and aggregates this information on a per variant basis, describing, among other things, the likely inheritance pattern while simultaneously incorporating external annotation. Furthermore, VASP uses parental allele segregation patterns to determine phasing of variants and identifies genomic blocks common to affected pedigree individuals. VASP and other tools that aggregate information across a pedigree are critical for successful causal variant detection as they automate a complex task too labor intensive to perform manually.

2.1 Input

2.1.1 Input files

Two input files are required; a pedigree (PED) file representing pedigree structure and a variant call format (VCF) file containing variant information. The third required argument is either a variant effect predictor (VEP) annotation file (McLaren et al., 2010) or the path to a local VEP installation when no annotation file is readily available. Optional input files include a text file with the path to individual binary sequence alignment/map (BAM) files used for determining zygosity and inheritance patterns. VASP is able to support any genome build providing all input files refer to the same reference genome.

2.1.2 Input filters

A key feature of VASP is flexible options with prioritization and ordering of variants based on stipulated criteria being specified at the outset. Filtering options include specific inheritance patterns, genomic region, gene name, population allele frequency, phasing information, number of affected/unaffected variant individuals and certain polyphen2 and sorting intolerant from tolerant (SIFT) categories. Combining parameters results in a dramatic reduction in the length of candidate variant lists—and the remaining variants will better correspond to hypotheses about the genetic basis of a particular disease.

2.1.3 Input variant set

Variant callers typically employ a quality score as a cutoff, above which lie a set of presumed high-quality variant calls. In reality, variants above this cutoff often include false positives, whereas variants below this cutoff include true positives (Weisenfeld et al., 2014). To minimize and potentially avoid this problem, the union of all variants called within the pedigree is used as the initial input variant set. This approach allows variant calls at a particular genomic position to be reconciled across the pedigree, potentially correcting calls lying near the cutoff for a single individual.

2.2 External variant annotation

Individual variants are annotated with VEP data including population frequency information from dbSNP (Sherry et al., 2001) and the 1000 genomes project (Genomes Project et al., 2010) as well as SIFT (Kumar et al., 2009) and PolyPhen2 (Adzhubei et al., 2010) scores for estimating the functional effect of missense mutations.

2.3 Pedigree-wide annotation

2.3.1 Genome phasing and de novo mutations

Whenever possible the parental allele inherited by each child is determined, and this information is clustered into genomic blocks of presumed shared inheritance to obtain genome-phasing information. Variants demonstrating segregation differences between affected and unaffected children are further prioritized. In addition, any variant exhibiting Mendelian inconsistencies is annotated, with special attention paid to putative de novo mutations (i.e. non-mutant, unaffected parents and a heterozygous offspring).

2.3.2 Disease inheritance patterns and compound heterozygosity

For each variant in the pedigree the inheritance is determined and annotated accordingly. The zygosity of particular variants is preferentially determined from raw sequence data obtained from BAM files using SAMtools (Li et al., 2009) but failing this the required genotype field (GT) and optional allele depth (AD) tags from the VCF file are utilized. Compound heterozygote genes are also annotated, defined as genes containing at least one heterozygous SNV or indel inherited from each parent with unaffected and affected siblings not sharing identical heterozygous variants. These variants must further be heterozygous in all affected individuals and not be homozygous in any unaffected individuals. These compound heterozygous genes are further prioritized in cases where each parent contributes rare or novel alleles.

2.3.3 Gene variability statistics

For each gene three measures of variability are reported; total number of variants, total number of unique variant coordinates and percentage of total transcript bases found variant. Increased gene variability may be relevant to particular diseases but also may be indicative of read alignment issues (often due the presence of gene duplicates) or may indicate the gene is functionally redundant and thus not functionally constrained.

2.4 VASP output and ordering

VASP reports contain all variants detected in at least one pedigree member and categorises variants as either novel, rare (0–2% population frequency), no frequency (known variant but no frequency data available) or common (>2% frequency). For each variant VASP reports both pedigree-wide information (such as inheritance pattern or phasing data) as well as variant-specific information (such as population frequency or polyphen score). By default VASP reports are sorted progressively on four measures: variant category (novel, rare, no frequency and common), the number of variant affected samples (in descending order), the number of unaffected variant samples and lastly the variant population frequency.

3 Results

VASP makes no assumptions regarding the underlying disease transmission mechanism, an apparent strength when compared with similar software (Supplementary Table S1). Instead, VASP provides powerful filters with the aim of allowing researchers to harness their additional knowledge of the disease to generate reduced variant lists suitable for manual interrogation. One current limitation of VASP is that it can only be run on the command line.

Five pedigrees (Supplementary Table S2) were analyzed to calculate variant segregation statistics with pedigree G1 (Supplementary Figure S1) variant lists (Supplementary Table S3) taken forward to illustrate the effect of various filtering strategies (Supplementary Table S4). To date VASP has been used to analyze 45 pedigrees and found strong candidate causal variants in 15 of these (33.3%). These 15 pedigrees exhibit a wide array of disease transmission mechanisms including autosomal dominant and recessive inheritance, de novo mutations, compound heterozygosity and more complex multi-gene cases. This variety in transmission mechanisms within this relatively small group sharing similar diseases illustrates the importance of flexible pedigree analysis software.

We present VASP, a flexible tool for identifying putative causal variants from pedigree sequence data. Through aggregation of data for genetic variants across pedigree members, VASP allows powerful, custom variation prioritization, taking advantage of external datasets and prior knowledge of disease incidence and inheritance patterns. With this tool users have the opportunity to greatly narrow the number of candidate causal variants, using custom criteria, to a size suitable for manual interrogation.

Funding

This work was supported by National Health and Medical Research Council Australia Fellowship 585490, National Institutes of Health [grant number U19 AI100627] and Bioplatforms Australia.

Conflict of Interest: none declared.