Genomic Predicting Single-Cross Maize Hybrids’ Grain Yield Stability: Is it an Efficient Strategy?
Zea mays; Genotypes-by-environments interaction; Genomic selection; Stability index; Bayesian Multi-Environment approach.
Tropical agriculture provides intense climatic challenges. Although its relevance, off-season maize presents greater crop risks primarily due to lower water availability. Furthermore, the genotypes-by-environments (G-by-E) interaction is even more challenging to maize breeding in tropical conditions. In this scenario, the identification of highly stable maize hybrids is a powerful strategy to minimize G-by-E effects. The genomic prediction (GP) approach has been increasingly employed in plant breeding, and modeling the G-by-E is an efficient strategy to improve prediction accuracy. However, modeling this interaction is computationally demanding and time-consuming. Therefore, simpler and easier strategies could be applied to perform GP to select highly phenotypic stable genotypes. The objective of this work was to identify simple and efficient strategies to perform the genomic selection by directly predicting single-cross maize hybrids' stability. Nine indices were estimated using a balanced date set of 185 maize hybrids evaluated over six environments. Genomic predictions were performed applying a Bayesian Ridge Regression model. Ten rounds of 10-fold cross-validation were performed to evaluate predictive ability (PA) and accuracy (ACC). Considering the nature of the data set, seven out of nine indices present satisfactory PA and ACC, fluctuating from 0.17 to 0.31, and from 0.33 to 0.61, respectively. On average, indices predicting outperform the Bayesian Multi-Environment approach in 29.7%, considering the seven well-predicted indices. Wricke’s ecovalence index and the main effect of genotypes present better performance, however, considering larger data sets, the Euclidian Distance and the Harmonic Mean of the Relative Performance of the Breeding Values (MHPRVG) indices should be more adequate.