Application of Genomic Models with Genotype-By-Environment Interaction to Maize and Wheat Hybrid Prediction and Line Cross-Prediction.

Genomic statistical model that incorporate G×E can use different kernels and, in general, multivariate genomic models with G×E increase prediction accuracy as compare with single-site model. Models for general and specific combining ability for predicting the performance of hybrids in environments have been developed. We have applied these models to an extensive hybrid maize population of hybrids derived from dent and flint lines, which were evaluated in 58 environments over 12 years. On average, genomic models that include the interaction of general and specific combining ability with environments have greater predictive ability than genomic models without interaction (ranging from 12-22%). The genomic estimated breeding value (GEBV) could be used for selection of parent from a breeding population with superior general combining ability. However, assessing specific combining ability required the identification of pair of inbreeds that in cross combine and complement well their cumulate favorable causal loci. Therefore, crosses with high Mendelian sampling variance supposed to generate progenies with high probability of accumulate favorable causal loci with ½ of additive variance cause by males and the other half caused by female. We investigate methods for cross-prediction based on the parental contribution and the Mendelian sampling from doubled haploid lines developed from a large number of in silico bi-parentals.