Scaling factor assessment in single-step GBLUP evaluations for small genotyped populations: A case study on Iranian Holstein cattle

Abstract

The estimation of genomic breeding values (GEBV) with the single-step genomic BLUP (ssGBLUP), which combines genomic (G) and pedigree relationship matrices in the single-step relationships matrix (H), may require the specification of two scaling parameters, τ and ω. In this study, we evaluate different scenarios to determine the best combinations of these scaling factors (τ for G−1 and ω for A22−1), and emphasize their relevance for single-step genome-wide association studies (ssGWAS). We use a dataset of about 3.1 million Iranian Holstein cattle, including 827,396 and 469,676 primiparous cows with records on milk yield (MY) and somatic cell score (SCS), respectively, and 2,189 genotyped bulls. The choice of the scaling factors was in particular relevant for dispersion (measured by regression coefficient b1 in a linear regression). Using default value s of τ=ω=1 resulted in a slight dispersion in SCS (b1=1.038) compared to MY (b1=0.88). Optimal combinations of scaling factors for each trait were selected based on the validation metrics, accuracy and b1, and the outputs of post genomic analysis (like ssGWAS). Concerning the latter, the behavior of P-values and especially prediction error variance (PEV) of SNP effects when changing τ and ω, provided helpful information complementing the metrics accuracy and b1. Overall, our results confirm the importance of using the τ, and ω parameters in both (ss)GBLUP and (ss)GWAS, and suggest that the combinations of (τ, ω) equal to (1.2, 0.7) and (1, 1) are the best choices for MY and SCS on the considered dataset.