Effects of marker density on genomic prediction for yield traits in sweet corn

By accounting for many traits, phenotyping sweet corn is a costly practice, making complementary strategies necessary. Thus, predictive methods present as an excellent alternative for the prediction and selection of the traits. The accuracy of the prediction is highly influenced by characteristics such as phenotypic data quality and marker density, which impact on project costs. Several studies have been carried out to verify minimum densities without the significant loss in prediction accuracies, but none with sweet corn. In this study, the objectives were to test, assess and validate different strategies to pre-select markers for genomic selection and to find the minimum density in a prediction of yield traits in sweet corn. Initially, the prediction was performed with a high-density chip and then, using a pre-selection strategy of clustering markers into haplotype blocks. Furthermore, a third strategy was tested, where markers were selected evenly across the genome. In general, all traits showed a significant reduction in accuracy as the number of markers decreased. However, the relationship between marker's increment and accuracy did not remain constant and reached a plateau after a certain point. Applying marker pre-selection can be a good option for a cost-efficient implementation of genomic selection in sweet corn for yield traits, as they can be predicted with a significant accuracy using a panel of similar to 8k quality markers that are evenly across the genome. Furthermore, using one marker per haplotype block appears to be a better cost-effective strategy for carrying out genomic selection in sweet corn, for yield traits.