Genotype-by-environment (GxE) interactions is a reality that scientists deal with when developing new and better varieties. With increase in scale of both phenotypic and genetic data, coupled with environmental data, prediction of environment-specific multi-environment trials (MET) is gaining importance. We leveraged phenotypic and genotypic data for ~150 clones and five checks evaluated in 31 environments (location-season-year combination) to define our mega environments. Further, we used this dataset to counter different prediction problems faced in cassava breeding such as (i) predicting for unobserved genotypes across environments, (ii) predicting for unobserved genotype in never evaluated environments, and (iii) making predictions for unobserved environments. No clear grouping of the environments was observed, based on the planting seasons or proximity of the trials from the phenotypic data of the five-checks. Our prediction accuracies for the three prediction strategies ranged from 0.47 in CV2 to 0.91 in CV3 for CBSDRs. From this study, we established that CBSD (assessments undertaken at three, six and at harvest) can be predicted with reasonable accuracies under different scenarios that mimic real problems encountered in cassava breeding.
Date of publication:
RUFORUM Working document series
Agris Subject Categories: