166
Use of Genomic Recursions and APY Algorithm for Single-Step GBLUP Analyses with Large Number of Genotypes
The purpose of this study was to examine accuracy of genomic selection in single-step genomic BLUP (ssGBLUP) when the inverse of the genomic relationship matrix (G) is derived by the algorithm for proven and young animals (APY). This algorithm implements the inversion of G by genomic recursions, with recursions for young animals involving only the proven animals. With efficient implementation, the algorithm has a cubic cost for proven animals but only a linear cost for young animals. Simulated data set included 142k phenotypes in 6 generations under selection for EBV, with 170k animals in the relationship matrix. Genomic data consisted of 20k animals genotyped for 45k SNP; the simulated genomic data mimicked the bovine genome. The proven animals were 10k genotyped parents selected from the first 5 generations, and the young 10k genotyped animals were selected from the last generation. For animals treated as young, 5k had a single record and 5k had no records. Comparisons involved GEBVs obtained by ssGBLUP evaluation with either the exact G (G-REG) and the G inverted by APY algorithm (G-APY). The correlations between GEBV with the G-REG and G-APY were 0.97 overall, 0.94 for animals treated was young without records, and 0.98 for animals treated as young with records. The true accuracies for the animals with records with G-REG and G-APY were 0.57 and 0.58, respectively; for the animals without records, the accuracies for REG and APY were both 0.43. When the status of the young and proven animals was switched, the accuracies remained identical. A separate analysis involved a national data set for final score in Holsteins. Out of 74,980 genotypes for bulls, 29,552 for bulls with daughters were treated as proven and 45,428 without daughters were treated as young. The correlations of GEBV obtained with the REG and APY algorithms were >0.99 for both groups of bulls. When the number of high-accuracy animals with genotypes is limited (< 100k), the APY algorithm may drastically reduce the cost of the ssGBLUP evaluation without affecting the accuracy. The APY algorithm may allow using all the available genotypes in one ssGBLUP analysis to reduce biases due to preselection of young animals.
Keywords: single step method, genomic selection, genetic evaluation