Genomic prediction in plant breeding using environmental covariates

This project is developing efficient linear mixed models for genomic prediction within global plant breeding programmes, with particular focus on informative models for genotype by environment interaction using environmental covariates. Plant breeders often consider genotype by environment interaction as an impediment to efficient selection, since genotypes have a different response in different environments. The appealing feature of using environmental covariates for selection, such as soil moisture and daily temperature, is that genotype by environment interaction becomes directly interpretable, and thence predictable. The approach developed in this project integrates environmental covariates within a special factor analytic framework, so that both predictable and repeatable forms of genotype by environment interaction are jointly modelled using known covariates and latent factors, respectively.

This provides plant breeders with forward predictions of any genotype into any environment, and thence enables genotype by environment interaction to be used as a resource to inform, rather than impede, efficient selection.