Robust and accurate estimation of paralog-specific copy number for duplicated genes using whole-genome sequencing

Abstract

The human genome contains hundreds of low-copy repeats (LCRs) that are challenging to analyze using short-read sequencing technologies due to extensive copy number variation and ambiguity in read mapping. Copy number and sequence variants in more than 150 duplicated genes that overlap LCRs have been implicated in monogenic and complex human diseases. We describe a computational tool, Parascopy, for estimating the aggregate and paralog-specific copy number of duplicated genes using whole-genome sequencing (WGS). Parascopy is an efficient method that jointly analyzes reads mapped to different repeat copies without the need for global realignment. It leverages multiple samples to mitigate sequencing bias and to identify reliable paralogous sequence variants (PSVs) that differentiate repeat copies. Analysis of WGS data for 2504 individuals from diverse populations showed that Parascopy is robust to sequencing bias, has higher accuracy compared to existing methods and enables prioritization of pathogenic copy number changes in duplicated genes.

Variant calling in polyploids for population and quantitative genetics

“…No other polyploid genotype callers, to my knowledge, account for allele bias. Emerging solutions to reducing genotyping error from poor read mapping include the modification of variant calling algorithms developed for CNVs (Layer et al, 2014;Prodanov and Bansal, 2022) or ancient DNA . For example, the software ancient DNA software, snpAD (Prüfer, 2018) , iteratively estimates genotype probabilities and r , the frequency at which the sequences are sampled from the reference allele at heterozygous sites, to account for reference bias.…”

Locityper: targeted genotyping of complex polymorphic genes

“…We split the background region into windows of fixed size based on the mean read length, and assign reads to windows based on the middle of the corresponding read alignments. Next, we count the number of primary read alignments assigned to each window (only first mates are counted to preserve window independence) 74 .…”

Short-read aligner performance in germline variant identification

“…When a read sequence has equivalent high-scoring mappings at two or more reference genome locations (each associated with MAPQ ≤ 3), an aligner may report the alternative mapping locations in SAM/BAM format in different ways: by emitting multiple records, each flagged as secondary mappings; by reporting one of the possible mapping locations as a primary mapping and listing the alternative locations in an optional field; or by randomly “choosing” one mapping location to report. But regardless of how multiple possible mappings are reported for a read, significant computational analysis is still needed to disambiguate them prior to variant discovery ( Prodanov and Bansal 2022 ).…”