Positive and Semi-Positive Lessons In Genomewide Selection and Genomewide Association Mapping In Maize

Lower costs of genotyping than of phenotyping in maize have made genomewide selection (or genomic selection) and genomewide association mapping feasible in breeding. I will review key lessons from applying genomewide selection and association mapping in maize breeding. First, the choice of genomewide selection versus association mapping in a breeding program depends on the genetic architecture of the trait. Second, genomewide selection is most advantageous when heritability is high in the training population but is low or zero in the test population. I will present several examples of how such situation can be achieved. Third, genomewide predictions are usually most accurate with a simple model that assumes that each marker accounts for an equal proportion of the total genetic variance and that ignores epistasis. Fourth, traits differ in their prediction accuracy even when the marker density, population size, and heritability are kept constant. Empirical data are therefore needed to determine which traits are the most predictable and which traits are not. Fifth, for finding marker-trait associations, models that incorporate genomewide background effects are superior to the K and QK models commonly used for association mapping. Sixth, QTL mapping in biparental crosses is still needed even with association mapping. I will conclude by presenting future needs in applying markers in maize breeding.