Application of a single-step SNP BLUP random regression model to test-day yields and somatic cell scores in German Holsteins

Abstract

Test-day yields or somatic cell scores have been evaluated with a multi-lactation random regression test-day model for dairy cattle breeds in Germany. The Liu-Goddard single-step SNP BLUP model directly estimates the effects of SNP markers together with all other model effects and has been demonstrated to be most efficient in handing large genotype data among all variants of the single-step model. The aims of this study were to test the implementation of the single-step SNP BLUP model to the test-day traits in German Holsteins and to investigate accuracy and bias of the genomic prediction. Approximately one million genotyped Holstein animals were jointly evaluated with c.a. 12 million dairy cows having test-day data. Pseudo-phenotype data of more than 138,000 Holstein bulls were integrated as a correlated trait to the national test-day yields or somatic cell scores. A genomic validation was conducted by removing test-day records in last four years for the national cows and truncating the youngest four birth years of the integrated bulls. The single-step model gave higher correlation of the SNP effect estimates between the full and truncated evaluation than the current multi-step model. In addition, regression coefficient of the SNP effect estimates from the full on the truncated evaluation was closer to 1 for the single-step model. Based on the results for the validation bulls we can draw a conclusion that the single-step model leads to neither an inflation nor a deflation of genomic prediction for the four test-day traits. No post-processing of GEBV of young animals would be needed for the genomic prediction in German Holsteins. The impact of selecting bulls with foreign daughters was investigated. We have found that removing genotype records of older bulls led to average Mendelian sample effects closer to zero for genotyped female animals.