In admixed populations, markers may be associated to different QTL depending on the origin of a given genomic segment. The goal of this study was to investigate if taking into account of the breed origin of alleles in a breed of origin genomic model (BOGM) can improve genomic predictions compared to a traditional genomic model (TGM) in admixture populations. Real genotype data of Danish Holstein and Jersey breeds were used as base populations to simulate F1 crosses. This was followed by simulating 5 discrete generations of random mating to achieve a highly admixed population. A single trait with heritability of 0.25 and 100 QTL was simulated on the genome. Three different scenarios were considered, where the QTL effects of two breeds were sampled from a multivariate normal distribution with correlation 1.0, 0.5, or 0.1. Accuracy and bias of models which were measured as correlation and regression coefficient between true and estimated genomic breeding values, respectively, were validated using each first three generations as training, and subsequent generations as test populations. The BOGM had higher accuracy than TGM with a correlation of 0.5 (0.455 to 0.719 vs 0.285 to 0.594) and the correlation of 0.1 (0.419 to 0.749 vs 0.172 to 0.576). In the scenario with identical QTL effects in the two breeds, the accuracy of TGM was higher than BOGM (0.548 to 0.610 vs 0.456 to 0.557). Prediction bias was only observed when the first generation was used for training and subsequent generations as test. In conclusion, accuracy of genomic prediction in admixed population can be increased by taking into account the breed of origin of alleles. Key words: Genomic prediction, admixture populations, breed-specific effects, validation

Keywords