GBLUP Outperforms Quantile Mapping and Outlier Detection for Enhanced Genomic Prediction

Area of Science:

• Plant breeding
• Genomics
• Quantitative genetics

Background:

• Genomic selection (GS) is crucial for accelerating crop improvement by predicting complex traits.
• Accurate prediction models are essential for efficient breeding programs.

Purpose of the Study:

• To compare the predictive accuracy of genomic best linear unbiased prediction (GBLUP) against quantile mapping (QM) and outlier detection methods.
• To evaluate the impact of data adjustments on genomic prediction performance.

Main Methods:

• Utilized 14 real plant genetics datasets.
• Assessed predictive accuracy using Pearson's correlation (COR) and normalized root mean square error (NRMSE).
• Compared GBLUP, QM-adjusted GBLUP, and four outlier detection techniques.

Main Results:

• GBLUP consistently demonstrated superior predictive accuracy, achieving an average COR of 0.65.
• GBLUP provided up to a 10% reduction in NRMSE compared to alternative methods.
• Quantile mapping showed marginal benefits only in datasets with skewed distributions.

Conclusions:

• Genomic best linear unbiased prediction (GBLUP) is a robust and reliable method for genomic prediction in plant breeding.
• The utility of quantile mapping is limited to datasets with significant distributional deviations.
• Outlier detection had minimal impact on GBLUP's predictive performance.