The Influence of Peak GWAS Associations on Genomic Prediction Accuracy

Some of the most important agronomic crop traits of interest are complex and thus governed by many genes of small effect. The statistical models typically used in a genome-wide association study (GWAS) and genomic selection (GS) quantify the contributions of genomic markers in linkage disequilibrium with these genes to trait variation. In general, the GWAS has been successful at identifying genomic regions containing markers with moderate to strong marker-trait associations. It is possible to incorporate markers tagging such GWAS signals into breeding programs through marker-assisted selection, where plants with favorable alleles at the peak GWAS signals are selected for the next cycle of breeding. In the absence of such signals, GS is typically effective at accurately predicting trait values. These two strategies have been used separately until recently, when the predictive ability of GS models that include peak associated markers from GWAS as fixed effect covariates was assessed. Theoretically, these models should be optimal for predicting traits that have several genes of large effect and many genes of smaller effect. We expand upon this work by evaluating simulated traits from diversity panels in maize using a Ridge Regression Best Linear Unbiased prediction (rrBLUP) model that included fixed effects. Upon completion of this work, we anticipate being able to rigorously quantify the ability of fixed effect covariates tagging peak GWAS signals to increase GS prediction accuracy in the rrBLUP model under a wide variety of genetic architectures and genomic backgrounds.