How to enviromically predict breeding genotypes as if they were commercial cultivars?

Abstract

Bridging the gap between how breeding genotypes perform in trials and how they might fare as commercial varieties or cultivars remains one of the enduring challenges in plant breeding, further compounded by the complexities of genotype-by-environment (GxE) interactions. This study implemented and evaluated an unstructured - uns - bivariate enviromic reaction-norm model capable of predicting breeding genotypes as cultivars by integrating genomic and environmental data. All model components were developed from scratch in R, ensuring full methodological transparency and reproducibility. The model jointly estimated genetic and residual variance-covariance matrices for experimental trials and commercial stands using a derivative-free restricted maximum likelihood (DF-REML) approach with the BOBYQA algorithm. Results demonstrated that employing a SNP-based genomic relationship matrix yielded biologically consistent estimates and enabled the construction of spatial recommendation maps for genotype deployment across diverse environments (e.g., in fields established by farmers or growers). These findings underscore the importance of modeling experimental versus commercial performance as distinct but genetically correlated traits, while also demonstrating the feasibility of implementing complex enviromic models without reliance on specialized software. This approach offers methodological advances for predictive breeding, supporting more efficient selection and reducing reliance on extensive multi-environment trials.