Bovine breed-specific augmented reference graphs facilitate accurate sequence read mapping and unbiased variant discovery

Abstract

Background The current bovine genomic reference sequence was assembled from a Hereford cow. The resulting linear assembly lacks diversity because it does not contain allelic variation, a drawback of linear references that causes reference allele bias. High nucleotide diversity and the separation of individuals by hundreds of breeds make cattle ideally suited to investigate the optimal composition of variation-aware references. Results We augment the bovine linear reference sequence (ARS-UCD1.2) with variants filtered for allele frequency in dairy (Brown Swiss, Holstein) and dual-purpose (Fleckvieh, Original Braunvieh) cattle breeds to construct either breed-specific or pan-genome reference graphs using the vg toolkit . We find that read mapping is more accurate to variation-aware than linear references if pre-selected variants are used to construct the genome graphs. Graphs that contain random variants do not improve read mapping over the linear reference sequence. Breed-specific augmented and pan-genome graphs enable almost similar mapping accuracy improvements over the linear reference. We construct a whole-genome graph that contains the Hereford-based reference sequence and 14 million alleles that have alternate allele frequency greater than 0.03 in the Brown Swiss cattle breed. Our novel variation-aware reference facilitates accurate read mapping and unbiased sequence variant genotyping for SNPs and Indels. Conclusions We develop the first variation-aware reference graph for an agricultural animal (10.5281/zenodo.3759712). Our novel reference structure improves sequence read mapping and variant genotyping over the linear reference. Our work is a first step towards the transition from linear to variation-aware reference structures in species with high genetic diversity and many sub-populations.

Reference flow: reducing reference bias using multiple population genomes

“…We call this a “rough” upper bound because, while the personalized reference is ideal in that it contains the correct variants, the accuracy of alignment is also affected by tool-specific heuristics. A true upper bound would be hard to obtain, so we settle for the rough upper bound provided by the personalized genome, as in previous work [ 17 , 27 ].…”

Genotyping common, large structural variations in 5,202 genomes using pangenomes, the Giraffe mapper, and the vg toolkit

“…For comparison, we mapped the same reads to GRCh37 with BWA-MEM. Replicating a previous approach 23 , we used and 24 to call variants, and filtered to high-confidence variants (root mean square read mapping quality >= 40 and depth >= 25) that were called as heterozygous for all mappers. For these variants, we found the fraction of reads supporting alternate vs. reference alleles at each indel length (Figure 5(A)).…”

Evidence for and localization of proposed causative variants in cattle and pig genomes

“…With the increasing number of livestock genome assemblies and versions, researchers might adopt pangenome references that catalogue structural diversity within a species (e.g. [ 126 , 127 ]) and representations of graph genomes that store such pan-genome information in a single data structure [ 128 ]. Graph genomes allow bioinformatic methods (e.g.…”