Single-step genomic evaluations with 570K genotyped animals in US Holsteins

Abstract

The objectives of this study were to implement and evaluate the “Algorithm for proven and Young” (APY) for inversion of the genomic relationship matrix (G) in single-step genomic BLUP (ssGBLUP). Phenotypic data included 11,626,576 final scores on 7,093,380 US Holsteins and genotypes were available for 569,404 animals. Daughter deviations for young genotyped bulls with no classified daughters in 2009 but with at least 30 classified daughters in 2014 were computed using BLUP with all the phenotypes and pedigrees. Genomic predictions (GEBV) were obtained by ssGBLUP using phenotypes up to 2009.  We calculated the G inverse with APY based on genomic recursions on a subset of “base” animals. We tested several subsets including 9,406 bulls with at least 1 daughter, 9,046 bulls and 1052 dams, 9,046 bulls and 7,422 classified cows, and random samples of 5,000, 10,000, 15,000, 20,000, and 30,000 animals. Validation reliability was calculated as R² with a linear regression of daughter deviations on GEBV for young genotyped bulls. The reliabilities were 0.39 with 5,000 randomly chosen base-animals, 0.45 with base-animals including bulls and cows, and 0.44 with the remaining subsets. Setting up the G inverse for all the genotypes with 10,000 base-animals took 1.3 hours and 57GB of memory. Genomic predictions with G inverse are accurate when the number of base animals is at least 10,000. Single-step genomic BLUP using the G inverse via APY is applicable to populations with a large number of genotyped animals.