Multivariate Prediction of Untested Single-Cross Hybrids Using Phenotypic and Genomic Information Jointly.

Up to now, the performance of untested single-cross hybrid has been predicted on the basis of either phenotypic or genomic information. In addition, when the specific combining abilities of the lines are predicted on the basis of only genomic information, the incidence matrix is the result of multiplying element by element the marker genotypes of the parental lines. This has produced erratic genomic hybrid prediction that has led to conclude that genomic prediction of hybrids is useless. A multivariate mixed model to predict performance of single-cross hybrids based on genomic and phenotypic information simultaneously is proposed. In addition, instead of creating the incidence matrix for the specific abilities with parental lines marker genotypes, this will be made with F1-population marker genotypes. This new incidence matrix must contain precise information of the F1 dominance deviation effects that can not occur with the homozygous-lines marker genotypes. Furthermore, with the joint information of two sources the precision of the prediction of the general and specific combining ability effects should be increased. The proposed model would be based on phenotypic information, parental-line information, parental-line marker genotypes, and F1-population marker genotypes. The multivariate proposed approach will improve the prediction of the parameters because the genotypic and genomic correlation between lines and traits are taken into account; that is not possible with the univariate approach.