Effect of using preselected markers from imputed whole-genome sequence for genomic prediction in Angus cattle

Abstract

The advent of next-generation sequencing enables the opportunity to use denser marker tools, up to whole-genome sequences (WGS), for genomic prediction in livestock. Improvement in genomic prediction (GP) accuracy from using WGS has been observed in simulation studies. In contrast, such advantage has found to be inconsistent once implemented in practice. The benefit of WGS appears to be from markers that are significant for the trait of interest. Thus, the main objective of this study was to investigate the predictive ability of adding preselected markers to the standard-industry 50k genotype for GP of economically important traits in Angus cattle, namely, birth weight (BW), scrotal circumference (SC), carcass weight (CWT) and carcass intramuscular fat (CIMF). Animals were genotyped with either commercial or customised SNP-genotyping arrays; then, the genotypes were imputed to WGS. The 50k genotype was used as the control group. Informative markers associated with the desired traits were extracted from WGS, then were added to the 50k genotype. Several methods were chosen to select different sets of informative markers, including LD-based pruning, top SNP from a genome-wide association study (GWAS), functional annotation based on Gene Ontology, cattle QTL database, and sequence annotation. In total, eight different sets of genotypes were investigated. We applied different statistical models to predict genomic breeding values, including GBLUP, BayesR, and BayesRC, and two-GRM GBLUP constructed separately from the 50k and the preselected genotype set. Heritability (h2) estimates were similarly calculated using different sets of genotypes and statistical methods across all traits. The log-likelihood ratio values revealed that two-GRM GBLUP was more suitable than the single-GRM GBLUP. There was no significant difference in accuracy and bias among the different sets of genotypes compared to the control group or the statistical methods, except for BW. For BW, the Bayesian models slightly outperformed GBLUP. The findings suggest that potential improvements may be achieved by using preselected SNPs from the GWAS, a method that has proven within the population. The performance of preselected markers on GP influenced by several factors, including population structure, method used to select significant markers, and genetic architecture of traits.