Regularized quantile regression applied to genome-enabled prediction of quantitative traits

Area of Science:

• Quantitative Genetics
• Statistical Genomics
• Animal Breeding

Background:

• Genomic selection (GS) utilizes whole-genome markers to predict genetic merits and enhance breeding value (BV) prediction accuracy.
• Traditional GS models often assume normal phenotypic distributions, limiting their effectiveness with skewed data common in complex traits.
• Existing methodologies inadequately address statistical challenges posed by non-normal phenotypic distributions in GS.

Purpose of the Study:

• To propose and evaluate a novel regularized quantile regression (RQR) approach for genomic selection (GS).
• To improve the estimation of marker effects and genomic estimated breeding values (GEBV) under non-normal distributions.
• To compare the performance of RQR against traditional Bayesian LASSO (BLASSO) under various distribution scenarios.

Main Methods:

• Simulated genomic data for 1000 individuals with 1500 markers, incorporating small and large effect markers.
• Evaluated symmetrical, positively skewed, and negatively skewed phenotypic distributions.
• Applied Bayesian LASSO (BLASSO) and regularized quantile regression (RQR) at quantiles 0.25, 0.50, and 0.75.

Main Results:

• Regularized quantile regression (RQR) demonstrated efficiency in estimating genomic estimated breeding values (GEBV).
• RQR outperformed BLASSO across all evaluated skewed distribution scenarios when an appropriate quantile was selected.
• Significant gains were observed: 86.28% for positively skewed and 55.70% for negatively skewed distributions compared to BLASSO.

Conclusions:

• Regularized quantile regression (RQR) provides a robust and efficient method for genomic selection (GS), especially with non-normal trait distributions.
• The RQR approach enhances the accuracy of marker effect estimation and genomic estimated breeding value (GEBV) prediction.
• This methodology offers substantial improvements over traditional models for breeding programs dealing with skewed phenotypic data.