Tuning indirect predictions based on SNP effects from single-step GBLUP

Abstract

The objectives of this study were to investigate whether SNP effects can be accurately estimated when the algorithm for proven and young (APY) is used in single-step GBLUP (ssGBLUP), and how close indirect predictions, based on SNP effects, are to GEBV from regular ssGBLUP. Tests involved an American Angus dataset with 8 million animals in the pedigree. Among those, 80 993 were genotyped. Validation animals (15 040) were born from 2013 to 2014. The reduced dataset had genotypes and phenotypes up to 2012; the complete dataset had genotypes up to 2014 and was used to obtain the benchmark GEBV. Based on the reduced dataset, GEBV were calculated using regular ssGBLUP with direct inversion of G (G^-1), and APY ssGBLUP () with 11 000 core animals. The SNP effects were calculated based on a) G^-1, b) , or c) the inverse of the core portion of G (). Direct genomic values (DGV) for validation animals were obtained as the sum of SNP effects weighted by the genotype content, and the difference between pedigree and genomic base was added to obtain indirect predictions. Correlations between SNP effects obtained with G^-1 and were > 0.99; the lower correlation (0.93) was observed when using . Correlations between the benchmark GEBV and DGV from G^-1, , and were all 0.99. The average difference between benchmark GEBV and DGV was 113.95, indicating a large bias. Indirect predictions that include DGV and the difference between pedigree and genomic base were less biased, and therefore, comparable to GEBV. Accurate indirect predictions can be obtained when APY ssGBLUP is used. Backsolving genomic predictions to SNP effects may require only a group of genotyped animals representing the dimensionality of the genomic information.