Genomic Selection for Yield and Seed Protein Content in Soybean: A Study of Breeding Program Data and Assessment of Prediction Accuracy

Abstract

Soybean [Glycine max (L.) Merr.] is a major crop with high seed protein content. Genomic selection is expected to be a valuable tool in improving the efficiency of breeding programs, especially for complex traits such as yield. This study aimed to evaluate the accuracy of genomic selection for yield and seed protein content in a soybean breeding population. Having a structured population, we compared genomic prediction accuracy obtained using models calibrated across or within two subpopulations: early lines and late lines. Calibrations within subpopulations were more efficient. Using a medium density of markers and genomic best linear unbiased prediction (GBLUP) model, which assumes an additive polygenic architecture, we predicted ∼32 and 39% of phenotypic variation among late lines for seed protein content and yield, respectively. Prediction accuracy was further improved by including epistasis in the GBLUP model. Further, we assessed accuracies obtained using several Bayesian models that assume different distributions for marker effects: Bayesian ridge regression, Bayesian LASSO, BayesCπ, and BayesR. Overall, these approaches did not improve prediction accuracy. In this study, we reported preliminary results relevant to the study of the efficiency of genomic selection use in a breeding program.