Daniel Tolhurst: Genomic prediction into future growing environments using environmental covariates

Daniel Tolhurst (University of Edinburgh, Roslin Institute) explained how to accurately model observable and predictable genotype by environment interactions for the use in plant breeding.

From Daniel:
This research introduces a new genomic selection approach which directly integrates environmental covariates within a special factor analytic framework. The factor analytic approach of Smith et al. (2001) is an effective method of analysis for multi-environment plant breeding datasets, but has limited biological interpretation since the underlying factors are unknown so the modelled genotype by environment interaction (GEI) is observable, rather than predictable. The advantage of using regressions on known environmental covariates, such as soil moisture and daily temperature, is that the modelled GEI becomes directly interpretable, and thence predictable. The approach developed in this research exploits the desirable features of both classes of model, and includes a predictive model for GEI using a joint set of known and unknown factors. This enables factor analytic models to be utilised for forward prediction into future growing environments, and thence provides global plant breeding programmes with an effective framework to improve genetic gain amid climate change.
The new factor analytic approach is demonstrated on a late-stage cotton breeding dataset from Bayer Crop Science.
The known factors (environmental covariates) explain 34.6% of the genetic variance across cotton growing environments in USA, which represents 92.7% of the crossover GEI. The unknown factors then explain 40.7% of the genetic variance, which represents 87.6% of the non-crossover GEI. 
Efficient selection of candidate genotypes is demonstrated using a recent set of factor analytic selection tools which involve measures of overall (mean line) performance and stability across observed and future environments.
The new approach improves the prediction accuracy of overall performance based on future environments by 80% compared to a traditional random regression.
The results demonstrate the ability of the new approach to accurately model observable and predictable GEI in a manner that is both informative and practical to plant breeding.

Publications:
https://www.researchsquare.com/articl...