Assessing genomic prediction accuracy for Holstein sires using bootstrap aggregation sampling and leave-one-out cross validation

Abstract Text:

Since the introduction of genomic prediction for dairy cattle in 2009, genomic selection has dramatically changed many aspects of the dairy genetics industry and enhanced the rate of response to selection for most economically important traits.  Young dairy bulls are genotyped to obtain their genomic predicted transmitting ability (GPTA) and reliability (REL) values.  These GPTA are the main factor in most purchasing, marketing, and culling decisions until the bulls reach five years of age and their milk-recorded offspring become available.  At that time, daughter yield deviations (DYD) can be compared with the GPTA computed several years earlier.  For most bulls, the DYD are similar to the initial predictions.  However, for some bulls, the difference between DYD and corresponding GPTA is quite large, and published REL are of limited value in identifying such bulls.  A method of bootstrap aggregation sampling (bagging) using genomic BLUP (GBLUP) was applied to predict the GPTA of 2,963, 2,963, and 2,803 young Holstein bulls for protein yield, somatic cell score (SCS), and daughter pregnancy rate (DPR), respectively.  For each trait, 50 bootstrap samples from a reference population comprised of 2011 DYD of 8,610, 8,405, and 7,945 older Holstein bulls were used.  Leave-one-out cross validation was also performed to assess the prediction accuracy when removing specific bulls from the reference population.  The main objectives of this study were: (1) to assess the extent to which current REL values and alternative measures of variability, such as the bootstrap standard deviation (SD) of predictions, could detect bulls whose daughter performance will deviate significantly from early genomic predictions and (2) to identify factors associated with the reference population that can cause inaccurate genomic predictions.  Correlations between bagged GBLUP predictions and 2014 DYD were lower than GBLUP predictions from the full reference population.  The SD of bootstrap predictions was a useful metric for identifying bulls whose future daughter performance may deviate significantly from early GPTA for protein and DPR.  Use of bootstrap predictions could prevent up to 50% of type I errors and roughly 10% of type II errors in sire selection decisions.  The removal of certain reference population bulls indicated that testing set predictions for protein were robust overall, but some bulls negatively affecting prediction accuracy were identified.

Keywords: genomic prediction, bootstrap sampling, dairy cattle