28 Genomic prediction for marbling score in Hanwoo cattle using sequence data

Abstract

Abstract As sequence information becomes available for some livestock species, there is a question on the amount of non-redundant information that may be embedded in the extra millions of SNPs or in causative variants within sequence. The objective of this study was to assess the gain in accuracy by using imputed whole-genome and selected SNP from sequence data in a beef cattle population. The dataset consisted of marbling score phenotypes for 545K Hanwoo cattle and pedigree information for 1.3M, of which 1,160 were genotyped for 50K SNPs. Imputation was done first to 777K and then to whole-genome sequence (WGS), which comprised 11,146,536 SNPs. Additionally, differentially expressed genes (DEG) and their 321,614 harboring SNPs were identified based on RNA-seq analysis of animals with high and low marbling score. An extra scenario combined 50K with DEG SNPs. Genomic predictions were obtained using GBLUP and single-step GBLUP (ssGBLUP) with and without weights, and BayesR. The last method could not be used for the WGS data because of the large number of SNPs. Predictive ability was calculated as the correlation between phenotypes adjusted for fixed effects and GEBV for 169 young animals. For all the methods, WGS and DEG had a slightly negative impact on predictive ability. Both GBLUP and BayesR had similar performances when using 50K, DEG, and 50K+ DEG, with predictive abilities equal to 0.19, 0.16, and 0.18, respectively. Predictive ability for ssGBLUP was 0.27, 0.26, and 0.27, in the aforementioned order. Using WGS, predictive ability for GBLUP was 0.17 and for ssGBLUP was 0.26. Weighting SNP differently did not improve predictions. As ssGBLUP uses all data available, not only genotyped animals with phenotypes as the other methods, it is more robust for genomic predictions. No gain in accuracy was observed, possibly because the selected sequence variants were not causative.