Classical and genomic prediction of hybrid sweet corn performance in organic environments

Abstract

AbstractPlant breeders need efficient systems to identify which inbreds to combine to create new hybrid cultivars. The North Carolina Design II (NC DII) is a useful mating design to evaluate the potential of hybrid varieties and their inbred parents. Genomic best linear unbiased prediction (GBLUP) models, either with or without the inclusion of a dominance term in the model, have been found to be an efficient method for using rich marker sets for prediction. This study used marker data and phenotypic data collected in 11 organic trials across six locations on 40 inbred sweet corn (Zea mays L.) genotypes and 100 hybrid progenies formed from four disconnected NC DII mating blocks to predict performance of untested sweet corn hybrids. In 2017, validation trials of 24 previously untested hybrids were grown in five organic environments to assess the correlation between actual performance and the performance predicted by GBLUP or NC DII general combining abilities (GCAs). Fivefold cross‐validation accuracy ranged from 0.29 to 0.82 for the GBLUP predictions based on additive effects alone (GBLUP‐A) and from 0.70 to 0.91 for GBLUP predictions based on combined additive and dominance effects (GBLUP‐AD). For all traits except flavor, addition of dominance effects to the model increased the cross‐validation accuracy. Correlations between values measured in the 2017 validation trials and values predicted from the 2015 and 2016 training trials ranged from 0.36 to 0.92 for GCA‐based predictions, from 0.34 to 0.94 for the additive GBLUP model, and from 0.38 to 0.92 for the additive plus dominance GBLUP model.