Genomic Prediction of Hybrid Wheat Performance

Accurate prediction of single cross performance is of fundamental importance to increase the selection gain in wheat (Triticum aestivum L.) hybrid breeding programs. We used experimental data from a commercial wheat breeding program as well as simulated data sets and evaluated the prospects of predicting grain yield performance of untested hybrids applying different cross-validation scenarios. We used ridge regression best linear unbiased prediction (RR-BLUP), BayesA, BayesB, BayesC, and BayesC pi facilitating genomic selection for additive and dominance effects. In total 90 hybrids were evaluated for grain yield in unreplicated trials at four locations. The parental lines were fingerprinted with a 9000 (9k) single nucleotide polymorphism array. We observed in the cross-validation study high prediction accuracies for all five genomic models with a slight superiority of RR-BLUP and BayesB. Interestingly, ignoring dominance effects resulted in equal or even higher prediction accuracies. This lack of improvement points toward the need for further refine the genomic selection models to more precisely estimate dominance effects.