Genomic prediction of genotype by environment interactions for wheat by coupling genetic and physiological modelling
Résumé
Genomic prediction (GP) models can be used to predict the performances of unphenotyped individuals thanks to their genotypic information. This approach was shown to be efficient for many species, but its interest in crops is limited by the presence of genotype x environment interactions (GEI): the ranking of the varieties often depends on the environments.
Recent studies proposed to adapt GP models to predict GEI by using environmental covariates. In the same way that molecular information is used to link the genotypes, the environmental covariates are used to link the environments. In these models environments with similar limiting factors are supposed to interact similarly with the varieties.
We propose here to evaluate and improve these approaches by using ecophysiological modelling. The prediction efficiency of different strategies were compared in a dataset comprising 220 elite wheat varieties, phenotyped for yield components in around 40 environments and genotyped with the TaBW420K SNP array within the BreedWheat project.
We show that the use of environmental covariates increased prediction accuracy in comparison to additive models, and that crop models were efficient to derive environmental covariates more relevant than those directly obtained with pedoclimatic information.