Non parametric vs. GBLUP model for genomic evaluation with large reference population in Holstein cattle

Abstract

In the last years, genomic selection has become an important component in dairy cattle breeding programs. Accordingly, different approaches are currently being developed, and used, to estimate genomic breeding values. The objective of this study was to compare the predictive ability of four methodologies to perform genomic evaluations in 25phenotypic traits (including productive, type and functional traits) using a large reference population of dairy cattle. The four evaluated approaches were Bayesian Reproducing Kernel Hilbert Spaces (RKHS), simple G-BLUP (GB), G-BLUP including a polygenic effect of 5% (GBP-5%) and G-BLUP including a polygenic effect of 10% (GBP-10%).The first two approaches use only genomic information, and the last two use both genomic and pedigree information. The data consisted on de-regressed proofs for 18,443genotyped bulls.A cross-validation was performed dividing the bulls into a training and a testingdata setborn before or after 2005, respectively. The results show that within the approaches using only genomic information, RKHS performs better than a simple GBLUP model. However, including polygenic effect improved GBLUP results. In general, RKHS performed slightly better, with larger predictive accuracy and lower mean square error, for the production traits, while GBP-5% performed better for type traits.Further research is needed to include pedigree information and to optimize the computational requirements of RKHS approach for routinely genomic evaluations.