(A)cross-breed Genomic Prediction

Calus, M.P.L.; Huang, H.; Wientjes, Y.C.J.; Napel, J. ten; Bastiaansen, J.W.M.; Price, M.D.; Veerkamp, R.F.; Vereijken, A.; Windig, J.J.


Genomic prediction holds the promise to use information of other populations to improve prediction accuracy. Thus far, empirical evaluations showed limited benefit of multi-breed compared to single reed genomic prediction. We compared prediction accuracy of different models based on two losely related and one unrelated line of layer chickens. Multi-breed genomic prediction may be successful when lines are closely related, and when the number of training animals of the additional line is large compared to the line itself. Multi-breed genomic prediction requires models that are lexible enough to use beneficial and ignore detrimental sources of information in the training data. Combining linear and non-linear models may lead to small increases in accuracy of multibreed genomic prediction. Multitrait models, modelling a separate trait for each breed, appear especially beneficial when elationships between breeds are very low, or when the genetic correlation between breeds is negative.