Impact of the complexity of genotype by environment and dominance modeling on the predictive accuracy of maize hybrids in multi-environment prediction models

The prediction accuracy of multi-environment prediction models can be affected by the complexity of the genotype by environment interaction (G×E). Moreover, depending on the trait genetic architecture, accounting for non-additive effects, such as dominance effects, may increase the prediction accuracy of genomic models. Hence, we aimed to verify empirically: (i) the impact of the genotype by environment complexity on the prediction accuracy of grain yield in maize hybrids; (ii) the advantage of dominance effects modeling for the prediction of maize hybrids in multi-environment trials; (iii) how parent information impacts on the prediction accuracy of hybrids in multi-environment genomic models. We used a dataset comprising 614 maize hybrids evaluated during two growing seasons, under two nitrogens regimes at two locations in Brazil. The prediction accuracies were obtained using four different validation systems (hybrids and half-sib families based sampling). Our results suggest that sampling entire half-sib families or individual hybrids can achieve similar accuracy estimates in multi-environment prediction models. Moreover, modeling dominance deviations in a multi-environment prediction model can significantly increase the prediction accuracy, mainly under high G×E complexity. Also, we found a linear relationship between prediction accuracy and G×E complexity. Furthermore, we observed significant increases in prediction accuracy of lowly correlated environments when information of a linking trial/environment was included in the prediction model.