Genomic Prediction of Kernel Water Content in a Hybrid Population for Mechanized Harvesting in Maize in Northern China

Abstract

Genomic prediction enables rapid selection of maize varieties with low kernel water content (KWC), facilitating the development of mechanized maize harvesting and reducing costs. This study evaluated and characterized the KWC and grain yield (GY) of hybrid maize in northern China and used genomic prediction to identify superior hybrid combinations with low kernel water content at maturity (MKWC) and high GY adapted to northern China. A total of 285 hybrids obtained from single crosses of 34 inbred lines from Stiff Stalk and Non-Stiff Stalk heterotic groups were used for genomic prediction of KWC and GY. We tested 20 different statistical prediction models considering additive effects and evaluating the impact of dominance and epistasis on prediction accuracy. Employing 10-fold cross-validation, it showed that the average prediction accuracy ranged drastically from 0.386 to 0.874 across traits and models. Eight linear statistical methods displayed a very similar prediction accuracy for each trait. The average prediction accuracy of machine learning methods was lower than that of linear statistical methods for KWC-related traits, but the random forest model had a high prediction accuracy of 0.510 for GY. When genetic effects were incorporated into the prediction model, the prediction accuracy for each trait was improved. Overall, the model with dominant and epistatic effects (G:AD(AA)) performed best. For the same number of markers, predictions using trait-specific markers resulted in higher prediction accuracy than randomly selected markers. When the number of trait-specific SNPs was set to 100, the prediction accuracy of GY increased by 33.27%, from 0.406 to 0.541. Out of all the 561 potential hybrids, the TOP 30 hybrids selected by genomic prediction would lead to a 1.44% decrease in MKWC compared with Xianyu335, a hybrid with a fast kernel water dry-down, and these hybrids also had higher GY simultaneously. Our results confirm the value of genomic prediction for hybrid breeding low MKWC suitable for maize mechanized harvesting in northern China. In conclusion, this study highlights the potential of genomic prediction to optimize maize hybrid breeding, enhancing efficiency and providing insights into genotype-accuracy relationships. The findings offer new strategies for hybrid design and advancing mechanized harvesting in northern China.