Regression approaches for modeling genotype-environment interaction and making predictions into unseen environments.

Area of Science:

• Agricultural science
• Biometrics
• Genetics

Background:

• Plant breeding increasingly uses environmental data for better predictions in new environments.
• Traditional methods like Finlay-Wilkinson regression are foundational but have limitations.
• Genotype-environment interaction is a key challenge in variety testing.

Purpose of the Study:

• To review and unify various linear mixed models for prediction in plant breeding.
• To demonstrate the close relationship between seemingly distinct regression methods.
• To present a new approach for estimating prediction uncertainty in new environments.

Main Methods:

• Review of linear mixed models, including factorial regression, Finlay-Wilkinson regression, reduced rank regression, and kernel/kinship-based approaches.
• Development of a common model-based prediction framework.
• Application of cross-validation and a novel model-based approach for uncertainty estimation.

Main Results:

• Environmental covariates significantly enhance prediction accuracy.
• Several distinct regression methods are shown to be closely related within a unified framework.
• A new method improves the estimation of prediction variance.

Conclusions:

• A unified model-based prediction framework simplifies and enhances genotype-environment interaction analysis.
• The proposed methods improve prediction accuracy and uncertainty assessment for plant variety testing.
• The findings are illustrated with a rice variety trial dataset, demonstrating practical applicability.