Bayesian variable selection for genome-wide association study of grain traits in rice.

Area of Science:

• Agricultural Science
• Genetics
• Bioinformatics

Background:

• Rice (Oryza sativa) is a vital global food staple with significant economic importance.
• Improving rice yield, stress tolerance, and grain quality through genetic research is a key agricultural goal.
• Genome-wide association studies (GWAS) are powerful for linking genetic variation to traits, but high-dimensional genomic data pose challenges.

Purpose of the Study:

• To compare the predictive performance of various frequentist and Bayesian modeling approaches for rice GWAS.
• To evaluate the effectiveness of different statistical models in identifying genetic markers associated with rice traits.
• To demonstrate the utility of Bayesian frameworks for accelerating genetic improvement in rice breeding programs.

Main Methods:

• Utilized rice genotype and phenotype data for comparative analysis.
• Applied multiple linear regression (OLS), LASSO, Ridge, Bayesian LASSO, BSLMM, and a Bayesian spike-and-slab prior model.
• Employed cross-validation with mean squared error and predictive correlation to assess model performance.

Main Results:

• The Bayesian spike-and-slab prior model generally outperformed classical frequentist methods.
• Superior prediction accuracy and effective variable selection were achieved with the spike-and-slab model.
• Bayesian methods demonstrated effectiveness in identifying informative genetic markers in rice.

Conclusions:

• Bayesian model selection frameworks offer significant advantages for plant GWAS and trait prediction.
• Bayesian approaches are effective for identifying informative markers, supporting marker-assisted selection in rice.
• These findings support the use of advanced statistical models to accelerate genetic improvement in crop breeding.