Improving reliability of genomic predictions for Jersey sires using bootstrap aggregation sampling

Abstract

Genomic selection has revolutionized the dairy genetics industry and enhanced the rate of response to selection for most economically important traits. All young bulls are now genotyped using commercially available single nucleotide polymorphism arrays to compute genomic predicted transmitting ability (GPTA) and reliability (REL) values. Decisions regarding the purchasing, marketing, and culling of dairy bulls are based on GPTA until roughly 5 yr of age, when milk-recorded offspring become available. At that time, daughter yield deviations (DYD) can be used to assess the accuracy of the GPTA computed several years earlier. Although agreement between predictions and DYD is often good, the DYD of some bulls differ widely from corresponding GPTA, and published REL are of limited value in identifying such bulls. A method of bootstrap aggregation sampling (bagging) using genomic BLUP (GBLUP) was implemented to predict the GPTA of 379, 379, and 342 young Jersey bulls for protein yield, somatic cell score, and daughter pregnancy rate, respectively. For each trait, 50 bootstrap samples from a reference population consisting of 2011 DYD of 1,738, 1,616, and 1,551 older Jersey bulls were used, and correlations between bagged GBLUP predictions and 2014 DYD were lower than GBLUP predictions derived from the full reference population. Although the bagged GBLUP approach did not improve the predictive correlations, it allowed computation of bootstrap predictive reliabilities across random samples of the reference population. The bootstrap predictive reliabilities could be a useful diagnostic tool for assessing genome-enabled prediction systems or evaluating the composition of a reference population. Our main objective was to determine if bagging GBLUP of young Jersey bulls could lead to measures of reliability that would be a useful alternative to published REL values. The standard deviations of bagged GBLUP predictions were found to weakly improve our ability to identify bulls whose future daughter performance may deviate significantly from early GPTA for protein, but not for somatic cell score or daughter pregnancy rate.