Genomic Selection Approaches for Improving Maize Productivity

Abstract

Genomic selection (GS) is a new breeding strategy to increase genetic gain for complex traits, like grain yield and water-limited yield in maize (Zea mays L.). Comparisons of nine GS models—genomic best linear unbiased prediction (GBLUP), Bayesian Ridge Regression (BRR), Bayes A, Bayes B, Bayes Cπ, Bayesian LASSO (BLASSO), Reproducing Kernel Hilbert Spaces (RKHS), Random Forest (RF) and Support Vector Regression (SVR)—using multiple predictive accuracy, genetic gain, computational efficiency, cross Bayes B stood out with the highest prediction accuracy under optimal irrigation and managed drought stress , and a 2.51 The cycle time was reduced from 4 to 1.4 years, and cost was reduced by 41.5%. Bayes B's prediction accuracy in testing environments was high (mean r = 0.782), demonstrating its generalisability. Training population gains were asymptotic (N = 250), and the prediction accuracy was not increased with higher marker density (18,500 SNPs) than this. GBLUP was fastest, but Bayes B's performance and stability compensated for its speed. Weighted performance scores were highest for Bayes B (91.7), followed by Bayes A (89.8) and Bayes Cπ (89.5). These findings suggest Bayes B is the optimal genomic selection method to boost maize productivity, and to breed new varieties fast, cheap and resilient to climate change