Genomewide prediction of tropical maize single-crosses

Genomic selection studies usually use few environmental models to predict progenies from biparental crosses. However, in a maize breeding program, numerous crosses need to be evaluated in multiple field trials to identify single-crosses with greater yield potential. In this study, 614 AFLP marker effects estimated in 250 maize single-crosses evaluated in 13 environments were used to assess the influence of training population size (N), number of markers (N M ), genotype-by-environment interaction (G × E) and population structure on the prediction accuracy (r MG ) for grain yield using RR-BLUP model. Cross validation analysis were performed within and across environments, and within and across groups of related single-crosses. Then, genomic selection was compared to phenotypic selection in identifying the best single-crosses in a maize breeding program scheme. In general, increasing the training population size and the number of markers did not led to higher accuracy estimates. Predicted accuracies from cross validation analysis within environments were significantly higher than between environments, indicating that the effect of G × E interaction was important. Accuracy estimates were also higher when training and validation sets were composed of related single-crosses. In all scenarios, wide intervals of accuracy were found, meaning that genomic prediction may not be effective depending on the single-crosses sampled. The use of genomic prediction in maize breeding programs was discussed emphasizing the need of a training set evaluated in multiple environments and designing a genomic selection experiment according to the population structure so as to reduce sample problems and maximize the accuracy and the success of prediction.