A robust Bayesian genome-based median regression model.

Area of Science:

• Genomics
• Statistical Genetics
• Animal Breeding

Background:

• Current genome-enabled prediction models often assume normally distributed errors, making them sensitive to outliers.
• Outliers in genetic or environmental data can reduce the accuracy of genomic prediction models.

Purpose of the Study:

• To propose a robust Bayesian genome median regression (BGMR) model for improved genomic prediction.
• To address the sensitivity of existing models to outliers by using a Laplace distribution for errors.

Main Methods:

• Developed a Bayesian genome median regression (BGMR) model fitting regressions to medians.
• Employed a Laplace distribution for error terms to enhance robustness against outliers.
• Utilized Markov Chain Monte Carlo sampling within a Bayesian framework.
• Compared BGMR performance against genomic best linear unbiased prediction (GBLUP) and Laplace maximum a posteriori (LMAP) methods using simulated and real genomic data.

Main Results:

• The BGMR model demonstrated higher prediction accuracies compared to GBLUP and LMAP when outliers were present.
• The proposed model effectively handled data with unknown outliers.

Conclusions:

• The BGMR model offers a robust alternative for genomic prediction, particularly in datasets with potential outliers.
• This approach can be valuable for breeders needing to predict and select genotypes accurately from data containing outliers.