Phenotype Prediction Under Epistasis.

Area of Science:

• Quantitative genetics
• Genomic prediction
• Statistical modeling in breeding

Background:

• Accurate phenotype prediction from genomic data is crucial for plant and animal breeding.
• Enhancing prediction accuracy requires advanced statistical models that capture complex genetic architectures.

Purpose of the Study:

• To introduce and evaluate three methods for genomic prediction: Genomic Best Linear Unbiased Prediction (GBLUP), Epistatic Random Regression BLUP (ERRBLUP), and selective Epistatic Random Regression BLUP (sERRBLUP).
• To compare the predictive ability of these methods using a wheat dataset and simulated phenotypes.

Main Methods:

• Genomic Best Linear Unbiased Prediction (GBLUP) models additive SNP effects.
• Epistatic Random Regression BLUP (ERRBLUP) incorporates all pairwise SNP interactions.
• Selective Epistatic Random Regression BLUP (sERRBLUP) incorporates a subset of pairwise SNP interactions selected by effect size or variance.

Main Results:

• sERRBLUP demonstrated a substantial increase in prediction accuracy compared to GBLUP and ERRBLUP.
• The improvement was most pronounced when an optimal proportion of SNP interactions, selected by effect size, was included.
• The EpiGP R-package efficiently processes large-scale genomic data for these methods.

Conclusions:

• Selective Epistatic Random Regression BLUP (sERRBLUP) offers superior prediction accuracy for genomic selection.
• The selection of SNP interactions based on effect size is a key factor for improving predictive performance.
• The EpiGP R-package provides a computationally efficient tool for implementing advanced genomic prediction models.