A Bayesian Genomic Multi-output Regressor Stacking Model for Predicting Multi-trait Multi-environment Plant Breeding...

Area of Science:

• Genetics and Breeding
• Statistical Genomics
• Machine Learning in Agriculture

Background:

• Accurate genomic-enabled prediction is crucial for plant and animal breeding programs.
• Existing multi-trait multi-environment models can be computationally intensive.
• There is a need for efficient yet accurate prediction models.

Purpose of the Study:

• To propose and evaluate the Bayesian multi-output regressor stacking (BMORS) model.
• To compare BMORS with univariate GBLUP and BMTME models for prediction accuracy and computational efficiency.
• To assess BMORS as an alternative for multi-trait, multi-environment genomic prediction.

Main Methods:

• Developed a two-stage BMORS model: univariate GBLUP with genotype × environment interaction (GE) in stage one, followed by Ridge regression with trait predictions as covariates in stage two.
• Compared BMORS predictions against univariate GBLUP and BMTME models using 7 maize and wheat datasets.
• Evaluated prediction accuracy and computational time for each model.

Main Results:

• BMORS achieved prediction accuracy comparable to univariate GBLUP and BMTME models.
• The BMTME model yielded the best prediction accuracy.
• BMORS demonstrated a significant computational advantage, being at least 9 times faster than BMTME.

Conclusions:

• The BMORS model is a viable and computationally efficient alternative for multi-trait, multi-environment genomic prediction.
• BMORS offers a practical solution for large-scale breeding programs requiring faster computation.
• Further research may explore optimizing BMORS for even higher prediction accuracy.