Curated variation benchmarks for challenging medically relevant autosomal genes

Abstract

The repetitive nature and complexity of some medically relevant genes poses a challenge for their accurate analysis in a clinical setting. The Genome in a Bottle Consortium has provided variant benchmark sets, but these exclude nearly four hundred medically relevant genes due to their repetitiveness or polymorphic complexity. Here we characterize 273 of these 395 challenging autosomal genes using a haplotype-resolved whole-genome assembly. This curated benchmark reports over 17,000 single nucleotide variations, 3,600 INDELs, and 200 structural variations each for human genome reference GRCh37 and GRCh38 across HG002. We show that false duplications in either GRCh37 or GRCh38 result in reference-specific, missed variants for short- and long-read technologies in medically relevant genes including CBS , CRYAA , and KCNE1 . When masking these false duplications, variant recall can improve from 8% to 100%. Forming benchmarks from a haplotype-resolved whole-genome assembly may become a prototype for future benchmarks covering the whole genome.

Improved structural variant discovery in hard-to-call regions using sample-specific string detection from accurate long reads

“…We then manually investigated the SVs that were exclusively called by , discovering that all them exhibited two alleles, one per haplotype (i.e., heterozygous SVs with two non-reference alleles). This result confirms previous findings [52] that heterozygous insertions in tandem repeats are among the most challenging classes of SVs to discover with current methods.…”

“…To perform a more thorough analysis of the HG002 individual, we considered the CMRG (Challenging Medically Relevant Genes) callset provided in [52] and we evaluated callers’ accuracy against it. The CMRG callset consists of 250 SVs falling in 126 challenging and medically relevant genes that were excluded from the previously published GIAB benchmark [65] due to their complexity: compound heterozygous insertions, complex variants in segmental duplications, and long tandem repeats.…”

“…Throughout this work we highlight the importance of accurate benchmarks of SV calling methods. We evaluated on a recent benchmark extensively curated over the HG002 sample [52] with the specific purpose of producing SVs occurring in genes of medical relevance, which are considered challenging for mapping-based and assembly-based SV prediction even from highly accurate long reads. This benchmark revealed that other methods fail to call heterozygous indels in highly homozygous regions or erroneous indels interpreted by a consensus approach.…”

“…Finally, assembly-based approaches first assemble the sequenced reads into longer contigs and use the assembled contigs to predict variants [49, 50, 51]. Assembly-based methods have recently been shown to provide superior performance to mapping-based tools [52].…”

“…Furthermore, the direct sequence (or the fixed-length k -mer) comparison performed in these type of tools can result in collapse of repeats and lower sensitivity/accuracy. Finally, assembly-based approaches are very computationally resource intensive and often require integration of data from multiple different technologies (i.e., long-reads, short-reads, and Hi-C) [55, 52], higher sequencing depths (35x was reported in [52]), and extensive polishing and post-processing to yield a high-quality de novo assembly suitable for variant prediction, thus making them impractical for large-scale SV discovery across many samples. Furthermore, these approaches usually represent a single hapoltype and might miss variants in the diploid genomes [56].…”