Adsa Logo White Adsa Title White

Genomic prediction with single-step genomic BLUP using a subset of genotypes in US Holstein.

Y. Masuda




Genomic prediction with single-step genomic BLUP using a subset of genotypes in US Holstein.
Y. Masuda*, S. Tsuruta, I. Misztal. University of Georgia Athens, GA.

As of January 2020, the US dairy database includes more than 3.8 million genotypes. Most of the genotypes are for heifers, and only a fraction of them have phenotypes. Although the use of all genotypes in genomic prediction is an ideal strategy, the same prediction-accuracy can be obtained using a subset of genotypes with a decrease of computing cost in single-step genomic BLUP (ssGBLUP). The objective of this study was to compare genomic predictions of young bulls between all genotypes and a subset of genotypes in US Holstein. We calculated the benchmarks using the full data set, provided by the Council on Dairy Cattle Breeding, including 61M phenotypes of 305-d protein yield, 36M pedigrees, and 2.3M genotypes. The benchmarks included daughter-yield-deviation (DYD) from pedigree BLUP, DYD from ssGBLUP, and GPTA from ssGBLUP. We cut off the last 4 years from the full data, and the truncated set included 841K genotypes up to 2014 (841K; ALL). Out of it, we created 2 sets of genotypes including bulls only (142K; BULL) and bulls and cows with records and with both parents known (256K; BULLCOW). For validation, we chose 3,250 bulls that had at least 50 phenotyped-daughters in 2018, but that had no daughters in 2014. The coefficient of determination (R2) and the slope coefficient (b1) were calculated from a linear regression of the benchmark on GPTA. For GEBV-ssGBLUP as the benchmark, R2 was 0.82 for ALL, 0.76 for BULL, and 0.82 for BULLCOW. Whereas b1 was around 0.90 in ALL and BULLCOW, BULL showed a lower value (0.83). We observed the same tendency in the other benchmarks. Using bull genotypes alone drop the accuracy probably because of limited information. The use of cow genotypes increases accuracy and reduces bias. The inclusion of genotyped heifers does not improve the accuracy of bull predictions. Statistics for validations based on GPTA-ssGBLUP indicate good stability of genomic predictions. Routine analyses by ssGBLUP can include only genotypes for bulls with daughters and cows, with the remaining animals predicted indirectly.

Keywords: genomic prediction, single-step GBLUP, selected genotypes.