Genomic prediction of maternal haploid induction rate in maize.

Genomic prediction (GP) might be an efficient way to improve haploid induction rate (HIR) and to reduce the laborious and time-consuming task of phenotypic selection for HIR in maize (Zea mays L.). In this study, we evaluated GP accuracies for HIR and other agronomic traits of importance to inducers by independent and cross-validation. We propose the use of GP for cross prediction and parental selection in the development of new inducer breeding populations. A panel of 159 inducers from Iowa State University (ISU set) was genotyped and phenotyped for HIR and several agronomic traits. The data of an independent set of 53 inducers evaluated by the University of Hohenheim (UOH set) was used for independent validation. The HIR ranged from 0.61 to 20.74% and exhibited high heritability (0.90). High cross-validation prediction accuracy was observed for HIR (r = 0.82), whereas for other traits it ranged from 0.36 (self-induction rate) to 0.74 (days to anthesis). Prediction accuracies across different sets were higher when the larger panel (ISU set) was used as a training population (r = 0.54). The average HIR of the 12,561 superior predicted progenies (μ ) ranged from 1.00-18.36% and was closely related to the corresponding midparent genomic estimated breeding value (GEBV). A predicted genetic variance (V ) of reduced magnitude was observed in the twenty crosses with highest midparent GEBV or μ for HIR. Our results indicate that although GP is a useful tool for parental selection, decisions about which cross combinations should be pursued need to be based on optimal trade-offs between maximizing both μ and V .