Canine hip dysplasia, a debilitating orthopedic disorder that leads to osteoarthritis and cartilage degeneration, is common in several large-sized dog breeds and shows moderate heritability suggesting that selection can reduce prevalence. Estimating genomic breeding values require large reference populations, which are expensive to genotype for development of genomic prediction tools. Combining datasets from different countries could be an option to help build larger reference datasets without incurring extra genotyping costs. Our objective was to evaluate genomic prediction based on a combination of UK and US datasets of genotyped dogs with records of Norberg angle scores, related to canine hip dysplasia. Prediction accuracies using a single population were 0.179 and 0.290 for 1,179 and 242 UK and US Labrador Retrievers, respectively. Prediction accuracies changed to 0.189 and 0.260, with an increased bias of genomic breeding values when using a joint training set (biased upwards for the US population and downwards for the UK population). Our results show that in this study of canine hip dysplasia, little or no benefit was gained from using a joint training set as compared to using a single population as training set. We attribute this to differences in the genetic background of the two populations as well as the small sample size of the US dataset.
Keywords: canine hip dysplasia; genomic best linear unbiased prediction; genomic selection; joint reference population; labrador retrievers.