Genomic prediction for stem rust resistance in the southern United States elite oat (Avena sativa L.) germplasm

Stem rust (SR), caused by Puccinia graminis f. sp. avenae , is a major threat to oat production in the southern United States, necessitating durable resistance to sustain productivity. Genomic prediction (GP) offers a promising approach to accelerate the development of resistant cultivars by leveraging genome-wide marker data to predict breeding values. In this study, we evaluated the accuracy of genomic prediction models for SR resistance using the Southern Oat Association Panel (SOAP), a 440-line multi-institutional panel genotyped with both a 3K SNP array and genotyping-by-sequencing (GBS). Field trials were conducted across seven environments between 2022 and 2024, and disease severity (SV) and infection response (IR) were assessed. Predictive ability (PA) was estimated under three cross-validation (CV) scenarios: CV1 (untested genotypes), CV2 (incomplete field trials), and CV0 (new environments). Across traits and scenarios, Bayesian models (BayesA, BayesB, and Bayesian LASSO) consistently achieved the highest PA, with GBLUP and RRBLUP performing nearly as well, while machine-learning methods (RF and GBR) were less effective. For IR, PA was highest under CV0 (~0.55), moderate under CV2 (~0.50), and lowest under CV1 (~0.45). For SV, PA reached ~0.62 in CV0, ~0.54 in CV2, and ~0.50 in CV1. Differences between the 3K and GBS platforms were minimal, indicating that both genotyping strategies provide sufficient coverage despite contrasting marker distributions and LD patterns. These findings highlight and demonstrate the potential of GS to reduce reliance on large-scale multi-environment phenotyping, enable prediction for new genotypes and environments, and accelerate genetic gain in oat breeding programs.