Single-step GBLUP using APY inverse for protein yield in US Holstein with a large number of genotyped animals

Abstract Text:

The objective of this study was to provide initial results in an application of single-step genomic BLUP with a genomic relationship matrix (G^-1_APY) calculated using the Algorithm of Proven and Young (APY) to 305-day protein yield for US Holsteins. Two G^-1_APY were tested; one was from 139,057 genotyped bulls with 12,895 core animals (APY140K) and the other one was from 764,029 genotyped animals with 12,913 core animals (APY760K). The predictor data set consisted of phenotypes recorded after 1989 and pedigrees limited to 3 generations back from recorded or genotyped animals. Genomic predictions (GPTA2011) were calculated for predicted bulls that had no recorded-daughters in 2011 but had at least 50 such daughters in 2015. We used the official daughter yield deviations published in 2015 (DYD2015) for the predicted bulls (N=3,797). We also used the official GPTA published in 2011 with a multi-step method as a comparison, although official methods have improved since then. Coefficient of determination (R²) and slope (b₁) were calculated from a linear regression of DYD2015 on GPTA2011. Using APY140K, the R² was 0.50 compared to 0.51 from the official GPTA. The b₁ was much better (0.98) compared with 0.81 from the official GPTA. With APY760K, the R² was 0.46 and b₁ was 1.08. Incorporating effect of a SNP related to DGAT1 increased R² to 0.51 for APY140K and 0.48 for APY760K. The decrease in R² with APY760K compared to APY140K could be due to inclusion of lower quality genotypes, or biases caused with the use of all genotypes with incomplete phenotypes. All the computations finished within 11 hours including 4.2 hours to set up APY-inverse with APY760K. Based on the linearity of the computation cost, using 1 million genotyped animals with the same model would require 14 hours of computations. Single-step GBLUP can provide genomic predictions for all genotyped bulls and cows while accounting for pre-selection. Further research will determine the impact of various factors affecting the reliability such as validation methodology, weighting SNP markers, and quality of genotyped data.

Keywords: genomic evaluation, Holstein, ssGBLUP