Genomic Bayesian functional regression models with interactions for predicting wheat grain yield using hyper-spectral...

Area of Science:

• Agricultural Science
• Genetics
• Data Science

Background:

• Modern agriculture utilizes hyperspectral cameras for crop trait prediction.
• Existing methods use selected wavelengths (vegetation indices), potentially missing information.
• Genotype × environment (G×E) and band × environment (B×E) interactions are crucial but often unaddressed in prediction models.

Purpose of the Study:

• To develop Bayesian functional regression models for predicting wheat grain yield using all hyperspectral bands.
• To incorporate G×E and B×E interactions into prediction models.
• To compare the accuracy and efficiency of these new models against conventional methods.

Main Methods:

• Proposed Bayesian functional regression models utilizing all available spectral bands.
• Implemented models using B-spline and Fourier bases.
• Incorporated genomic or pedigree information, main effects of lines and environments, and interaction effects (G×E, B×E).
• Evaluated models on 976 wheat lines across three environments using 250 spectral bands (392-851 nm).

Main Results:

• Models incorporating B×E interaction terms demonstrated the highest prediction accuracy for grain yield.
• Bayesian functional regression models and conventional models showed similar prediction accuracy.
• Genomic or pedigree information did not significantly increase prediction accuracy in this context.

Conclusions:

• Bayesian functional regression models with B×E interactions offer a highly accurate approach for predicting grain yield.
• Functional regression models are more parsimonious and computationally efficient than models using all individual bands.
• The study highlights the importance of considering spectral-environment interactions for improved crop yield prediction.