A unified haplotype-based method for accurate and comprehensive variant calling

Abstract

Haplotype-based variant callers, which consider physical linkage between variant sites, are currently among the best tools for germline variation discovery and genotyping from short-read sequencing data. However, almost all such tools were designed specifically for detecting common germline variation in diploid populations, and give sub-optimal results in other scenarios. Here we present Octopus, a versatile haplotype-based variant caller that uses a polymorphic Bayesian genotyping model capable of modeling sequencing data from a range of experimental designs within a unified haplotype-aware framework. We show that Octopus accurately calls de novo mutations in parent-offspring trios and germline variants in individuals, including SNVs, indels, and small complex replacements such as microinversions. In addition, using a carefully designed synthetic-tumour data set derived from clean sequencing data from a sample with known germline haplotypes, and observed mutations in large cohort of tumour samples, we show that Octopus accurately characterizes germline and somatic variation in tumours, both with and without a paired normal sample. Sequencing reads and prior information are combined to phase called genotypes of arbitrary ploidy, including those with somatic mutations. Octopus also outputs realigned evidence BAMs to aid validation and interpretation.Haplotype-based approaches have emerged as the method of choice for calling germline variants because these methods are robust to alignment errors from read mappers and have better signal-to-noise characteristics than positional approaches 1-7 . However, existing haplotype-based variant callers have several limitations. First, existing tools are sub-optimal for many problems as most implement models that assume either diploidy 1-3 or constant copy number 4-6 , and assume that samples are selected from an idealized population of unrelated individuals. Such models are appropriate for calling germline variants in small cohorts, but provide a poorer fit to data generated in other experimental designs, such as studies involving samples with known relatedness such as paired tumours, single-cells, and parent-offspring trios, or in pooled tumour and bacterial sequencing where samples are often heterogeneous. These limitations cause researchers to implement custom pipelines that may integrate various callers and involve post hoc filtering and interpretation [8][9][10][11][12][13][14][15][16] . Second, existing haplotype-based methods suffer from windowing artifacts as variants are evaluated in independent non-overlapping regions. This can lead to false calls in complex regions where reads support variants that fall outside the region being evaluated. Third, existing methods do not make a clear distinction between the haplotype sequence supported by the read data, and the mutation events that gave rise to it. This makes it challenging to assign appropriate prior probabilities to these haplotype sequences, because different sets of mutations can have very different biological pla...

Oocytes can efficiently repair DNA double-strand breaks to restore genetic integrity and protect offspring health

“…To identify mutations characteristic of gamma-irradiation-induced DNA damage, we restricted our analysis to identifying haploid mutations that occurred in only one individual and did not overlap annotated mouse single-nucleotide polymorphisms (SNPs) or repeats. Reads were mapped to the GRCm38 genome with Bwa mem 0.7.15 (14), and individual samples were called for mutations with Octopus 0.5.2 (15). Unique mutations were identified and overlapped with SNPs and repeats with Bedtools 2.28 (16).…”

An Extensive Sequence Dataset of Gold-Standard Samples for Benchmarking and Development

“…In order to understand the impact of choices for instrument, preparation, and coverage, we performed variant calling with several different widely-used, open-source methods: GATK4 [11], Strelka2 [29], Octopus [30], and DeepVariant v1.0 [12]. We assessed the accuracy of the pipelines against the Genome in a Bottle truth sets for HG001-HG007 [1].…”

Somatic variant analysis of linked-reads sequencing data with Lancet

“…Local assembly approaches have become the preferred solution to increase variant calling power compared to read alignment-based methods (Narzisi et al, 2014;Cooke et al, 2018;Chen et al, 2016;Wala et al, 2018;Rimmer et al, 2014;Mose et al, 2014;Li et al, 2013;Narzisi et al, 2014;Cooke et al, 2018;Chen et al, 2016;Wala et al, 2018;Rimmer et al, 2014;Mose et al, 2014;Li et al, 2013;Benjamin et al, 2019). Lancet is a somatic variant caller that leverages local assembly and joint analysis of tumor-normal paired data using region-focused colored de Bruijn graphs, with on-the-fly repeat composition analysis and a self-tuning k-mer strategy.…”