Date of Award
Master of Science
Rohan L. Fernando
Recent simulation studies have shown that genomic selection (GS) is an appealing method to select purebreds for crossbred performance. In the case of crossbred records, SNP effects could be estimated using an additive model, or a breed-specific allele model (BSAM). In most studies, either additive gene action, perfect knowledge of SNP effects or both has been assumed. It has been argued that dominance is the likely genetic basis of heterosis, therefore explicitly including dominance in the GS model might be beneficial to purebred selection for crossbred performance. Further, in beef cattle, the commercial populations often consist of admixtures of crossbreds with unknown breed composition. It is hypothesized that if the model includes dominance, breed differences in crossbreds will be explained by the SNP and therefore admixture can be ignored.
In this study, crossbreeding was simulated under a dominance model. The simulated genome consisted of 1,000 SNP and 100 QTL on a single chromosome of 100 cM. The dominance variance and heterosis were first chosen to be large enough to clearly detect any advantage of including dominance in the model, and then restricted to a more realistic setting to verify if the advantages would still hold. The performance of GS using the additive model, BSAM and the dominance model for training was evaluated based on 1) response to 20 generations of selection in a two-way crossbreeding program, and 2) accuracy of prediction in a variety of training populations. The influence of admixture on GS was investigated by either ignoring or considering breed composition in the model when admixed populations were used for training.
Results show that without retraining, the dominance model steadily gave greater response during selection, accumulating to an advantage of 32.2% over the additive model and of 29.2% over BSAM by generation 20. Extra response was the result of an increase in heterosis but at some cost to improvement of purebred performance. The rate of decline in accuracy during selection was less with the dominance model versus either BSAM or the additive model because SNP allele substitution effects were recomputed with updated allele frequencies every generation under the dominance model. For various training populations, the dominance model resulted in accuracies by 1.1% to 8.1% higher than the additive model, and by 2.8% to 23.7% higher than BSAM. In particular, the dominance model raised the accuracy when the training population was not the target population to be improved and/or the training population was less related to the validation population for selection. Under the more realistic parameters, the advantages of the dominance model still held, although in a reduced amount. When training was on an admixed population, including breed composition in any model did not improve accuracy of prediction.
In conclusion, when dominance gene action is present, using a dominance model for GS would result in a greater response to selection in purebred animals for crossbred performance. The dominance model allows GS over generations or across breeds using one set of estimates of the additive and dominance effects with a higher accuracy than either BSAM or the additive model. Further, breed composition can be ignored in GS even when an admixed population is used for training.
Zeng, Jian, "Genomic selection of purebred animals for crossbred performance under dominance" (2011). Graduate Theses and Dissertations. 10474.