Multi-platform discovery of haplotype-resolved structural variation in human genomes

Abstract

The incomplete identification of structural variants (SVs) from whole-genome sequencing data limits studies of human genetic diversity and disease association. Here, we apply a suite of long-read, short-read, strand-specific sequencing technologies, optical mapping, and variant discovery algorithms to comprehensively analyze three trios to define the full spectrum of human genetic variation in a haplotype-resolved manner. We identify 818,054 indel variants (<50 bp) and 27,622 SVs (≥50 bp) per genome. We also discover 156 inversions per genome and 58 of the inversions intersect with the critical regions of recurrent microdeletion and microduplication syndromes. Taken together, our SV callsets represent a three to sevenfold increase in SV detection compared to most standard high-throughput sequencing studies, including those from the 1000 Genomes Project. The methods and the dataset presented serve as a gold standard for the scientific community allowing us to make recommendations for maximizing structural variation sensitivity for future genome sequencing studies.

An open resource of structural variation for medical and population genetics

“…While the number of fully resolved SVs per genome in gnomAD-SV using the integration of multiple algorithms here (n = 7,439 SVs per genome from ~32X coverage WGS) is roughly twice that of existing references from short-read WGS, such as the 1000 Genomes Project (3,441 SVs from ~7X WGS) and the GTEx project (3,658 SVs from ~50X WGS), 1,42 it is lower than estimates from recent long-read WGS analyses (24,825 SVs from ~40X long-read WGS). 19 The technology and methods used here are thus blind to a disproportionate fraction of repeat-mediated SVs, and underestimate the true mutation rates within these hypermutable regions. Similarly, high copy state MCNVs often require specialized algorithms and manual curation to fully delineate their complicated haplotype structures, 12,65,66 suggesting that the 1,055 MCNVs reported here are an incomplete portrait of extreme copy-number polymorphisms.…”

“…Finally, we leveraged matched long-read WGS data available for four individuals to perform in silico confirmation of our SVs predicted from short-read WGS. 19,27,28 These analyses yielded a confirmation rate of 94.0% for SVs with breakpoint-level read evidence (92.8% of all SVs), and revealed that 59.8% of breakpoint coordinates from the gnomAD-SV callset were accurate within a single nucleotide of the long-read data, while 75.9% were accurate within ±10bp. In conclusion, despite the limitations of short-read WGS, the seven benchmarking approaches we applied here suggest that these data conform to many fundamental principles of population genetics, including Mendelian segregation, Hardy-Weinberg equilibrium, population stratification, and linkage disequilibrium, and that gnomAD-SV is sufficiently sensitive and specific to provide a contemporary resource for most applications in human genomics.…”

“…We used Pacific Biosciences (PacBio) long-read WGS data available for four samples in this study to perform in silico confirmation to estimate positive predictive value (PPV) and breakpoint accuracy for SVs in gnomAD-SV (Supplementary Figure 10). 19,27,28 Data Figure 1); for the purposes of visualization, the y-axis for all panels presented here has been restricted to a maximum of three interquartile ranges above the third quartile across all samples for each category.…”

“…In contrast to the established gold-standard methods for profiling SNVs and indels from WGS data, like the Genome Analysis Toolkit (GATK), 17 the uniform detection of SVs from short-read WGS has presented a much greater challenge. 18,19 Analyses of SVs require specialized computational methods that consider multiple SV signatures, and even high-coverage short-read WGS fails to capture a component of the variant spectrum accessible to more expensive niche data types such as long-read WGS, optical mapping, or strand-specific sequencing. 18,19 Current population references of SVs from WGS are thus restricted to the 1000 Genomes Project (N=2,504; 7X sequence coverage) or smaller European-centric cohorts.…”

“…18,19 Analyses of SVs require specialized computational methods that consider multiple SV signatures, and even high-coverage short-read WGS fails to capture a component of the variant spectrum accessible to more expensive niche data types such as long-read WGS, optical mapping, or strand-specific sequencing. 18,19 Current population references of SVs from WGS are thus restricted to the 1000 Genomes Project (N=2,504; 7X sequence coverage) or smaller European-centric cohorts. 1,20 These references are dwarfed by contemporary resources for coding SNVs and indels such as the Exome Aggregation Consortium (ExAC) and its second iteration, the Genome Aggregation Database (gnomAD), which have jointly analyzed >140,000 exomes.…”