Genomic selection using principal component regression.

Area of Science:

• Genomics
• Statistical Genetics
• Bioinformatics

Background:

• Genomic selection (GS) uses whole-genome SNP data to predict genetic values for quantitative traits in plants and animals.
• High-dimensional genomic data presents significant computational challenges for GS due to numerous predictors and fewer subjects.
• Principal Components Regression (PCR) and Partial Least Squares Regression (PLSR) are dimensionality reduction techniques used to address these challenges.

Purpose of the Study:

• To compare the performance and computational efficiency of PCR and PLSR in genomic selection.
• To demonstrate the advantages of PCR, particularly when combined with the HAT method, over PLSR.

Main Methods:

• Dimensionality reduction using Principal Components Regression (PCR) and Partial Least Squares Regression (PLSR).
• Application of the HAT method to PCR to circumvent traditional cross-validation.
• Evaluation of predictive ability using simulated and real-world agronomical, metabolomic, and transcriptomic trait data in a rice population.

Main Results:

• PCR demonstrated superior performance compared to PLSR in cross-validation scenarios.
• The HAT method significantly improved PCR's computational efficiency by eliminating the need for mandatory cross-validation, unlike PLSR.
• PCR's benefits were validated across various trait types, including agronomical, metabolomic, and transcriptomic data.

Conclusions:

• PCR, especially with the HAT method, provides a computationally efficient and effective approach for genomic selection.
• PCR offers advantages over PLSR in handling high-dimensional genomic data for trait prediction.
• The findings support the broader application of PCR in genomic selection for diverse biological traits.