Germline variant calling using the seven bridges graph toolkit (SBGT)

The human reference genome lays the basis for genomic analyses by enabling alignment of sequenced reads. However, the linear human reference genome released by the Genome Reference Consortium only represents a single consensus haplotype, and therefore constitutes a suboptimal representation of human genetic variation. Directed acyclic graph reference genome representations that incorporate information on genetic variation have been shown to improve the accuracy of read alignment, variant calling and other subsequent genomic analyses. A set of bioinformatics tools utilizing graph genomes (Seven Bridges Graph Toolkit; SBGT) for the analysis of next-generation sequencing (NGS) data has recently been published by Seven Bridges in preprint*. In this study, we benchmark the SBGT pipeline on germline variant calling and compare it against other state-of-the-art whole genome pipelines, namely BWA-MEM + GATK4 by the Broad Institute. The analysis includes five whole-genome samples (HG001-HG005) with the truth data established by the Genome in a Bottle Consortium. The results show that the SBGT pipeline has the highest accuracy in small (<50bp) INDEL calling with an F1-score of 0.25% above the average, while scoring 0.06% above the average in SNP calling. Moreover, we show that, unlike the other pipelines, the SBGT pipeline is capable of finding INDELs as long as several kilobases without any additional processing steps. The Mendelian inheritance discordance measurements on CEPH/CEU and Ashkenazim trios demonstrate that the SBGT pipeline offers the most consistent germline variant calling with a score that is 1-2 percentage points better than the other pipelines. Finally, we present an optimized version of the SBGT pipeline that can process a whole-genome sample with 30x coverage on the cloud in 2 hours and under $5.