Modeling Epistasis in Genomic Selection.

Area of Science:

• Genetics
• Bioinformatics
• Plant Breeding

Background:

• Genomic selection (GS) aims to predict breeding values using genomic information.
• Modeling epistasis (gene-gene interactions) in GS is computationally intensive.
• Extended genomic best linear unbiased prediction (EG-BLUP) and reproducing kernel Hilbert space regression (RKHS) are efficient alternatives.

Purpose of the Study:

• To prove the equivalence between EG-BLUP and other GS methods explicitly modeling epistasis.
• To elucidate how RKHS with a Gaussian kernel captures epistatic effects.
• To compare the prediction accuracy of different GS approaches using real data.

Main Methods:

• Theoretical equivalence proof between EG-BLUP and GS models with epistasis.
• Mathematical demonstration of RKHS Gaussian kernel capturing epistasis.
• Comparative analysis of GS methods on wheat and maize experimental datasets.

Main Results:

• EG-BLUP is equivalent to GS models that explicitly include epistasis.
• RKHS with a Gaussian kernel effectively models marker epistasis.
• Epistasis modeling enhanced prediction accuracy in selfing species (wheat) but not outcrossing species (maize).

Conclusions:

• Computational burden of epistasis modeling in GS can be reduced using EG-BLUP and RKHS.
• Epistasis modeling is beneficial for genomic prediction in selfing crop species.
• The utility of epistasis modeling depends on the mating system of the species.