Genomic Prediction Paves Way for the Identification of Multiple Rust Resistant Genotypes in Bread Wheat (Triticum...

Over the past two decades, genomic prediction (GP), also known as genomic selection (GS), has been widely adopted in plant and animal breeding programs worldwide. GP is a promising approach that utilizes genomic markers to estimate genomic-estimated breeding values, facilitating the selection of superior individuals. In this study, we assessed the performance of five GP models-genomic best linear unbiased prediction (GBLUP), reproducing kernel Hilbert space regression (RKHS), and Bayesian methods (BayesA, BayesB, and BayesC)-for predicting seedling resistance and adult plant resistance (APR) to leaf/brown rust (LR), stem/black rust (SR), and yellow/stripe rust (YR) in wheat using a panel of 347 diverse germplasm accessions. At the seedling stage, the BayesB model showed consistently high performance, particularly for LR and SR datasets, whereas all models showed relatively poor accuracy for YR. For APR, both GBLUP and BayesB performed comparably well, especially for LR, whereas all models exhibited lower predictive ability for SR and YR. Furthermore, GWAS-guided GP analysis revealed that intermediate SNP densities (100 to 500 markers) significantly outperformed the full marker set across all rust types and stages, with prediction accuracy in some cases doubling or more. These findings suggest that the use of a full marker set may introduce noise and reduce efficiency, whereas selection of top-ranked GWAS-based markers enhances predictive power. Additionally, favorable allele analysis identified two promising lines-CRP-165/42 and HGP1-435-with broad-spectrum resistance at both seedling and adult stages, making them a valuable source for future rust-resistance breeding programs.