Genomic prediction of bovine leukosis incidence in a US Holstein population

摘要

Bovine leukosis (BL) is a dairy cattle disease with a significant negative impact on several economically important traits such as milk yield, fertility and survival of animals. As there is no treatment or vaccine for this disease, finding a possible genetic solution to alleviate the problem would be extremely beneficial. Like other complex traits, the utilization of molecular marker information within a genomic selection approach might help in making selection decisions and potentially reduce the prevalence of the disease in dairy herds. However, the choice of the appropriate response variable and the statistical model are required for a successful genomic prediction program. As such, the objective of this study was to assess the prediction quality of genomic selection models for BL incidence. Milk Enzyme-Linked Immunosorbent Assay (ELISA) data obtained from a US Holstein population were analyzed using two modeling approaches: Bayes A and Bayes B, and three alternative response variables (i.e. pseudo-phenotypes): estimated breeding value (EBV), deregressed proofs free of parent average (DRP), and DRP with parent average added back after deregression (DRP_PA). The prediction ability for each combination of model and pseudo-phenotype was assessed based on the reliability, which was calculated as the square Pearson correlation coefficient between the response variable and the estimated genomic breeding values. Furthermore, to assess potential bias of predictions, response variables were regressed on their corresponding estimated genomic breeding values. Bayes A and Bayes B showed similar results across the three response variables analyzed in this study. Using DRP_PA yielded 6% and 10% higher prediction ability compared to EBV and DRP, respectively. In addition, DRP_PA revealed lower bias estimates. Genomic selection can be potentially applied for BL incidence to reduce the prevalence of the disease in dairy cattle herds. Adding back parents average to DRP may increase the reliability and reduce the bias of genomic selection for this trait.