A Multiple-Trait Bayesian Lasso for Genome-Enabled Analysis and Prediction of Complex Traits

Area of Science:

• Quantitative genetics
• Statistical genomics
• Bioinformatics

Background:

• Genomic prediction of quantitative traits relies on accurate statistical models.
• Existing methods like GBLUP and Bayesian LASSO have limitations in handling multiple traits simultaneously.
• Developing advanced methods is crucial for improving prediction accuracy in breeding programs.

Purpose of the Study:

• To introduce and evaluate a novel multiple-trait Bayesian LASSO (MBL) method for genome-based analysis and prediction.
• To compare the performance of MBL against established methods like GBLUP and Bayesian Cπ.
• To demonstrate the utility of MBL in real-world genomic datasets.

Main Methods:

• Developed a multivariate linear Bayesian regression model with a T-variate Laplace prior for regression coefficients.
• Utilized a Markov chain Monte Carlo sampling scheme for parameter estimation.
• Applied MBL to two distinct datasets: wheat grain yield across environments and Pinus tree traits (rust bin, gall volume).
• Benchmarked MBL against bivariate GBLUP and bivariate Bayes Cπ using a training-testing layout.

Main Results:

• MBL demonstrated differential shrinkage of marker effects, outperforming GBLUP in this aspect.
• For wheat data, MBL showed comparable prediction accuracy to bivariate GBLUP and Bayes Cπ.
• In Pinus data, MBL achieved superior prediction accuracy compared to bivariate GBLUP and single-trait Bayesian LASSO.
• The MBL method is implemented in the Julia package JWAS.

Conclusions:

• MBL provides a valuable new tool for genome-enabled prediction of complex traits.
• The method offers improved predictive performance, especially in scenarios with correlated traits.
• MBL expands the toolkit available for quantitative geneticists and breeders seeking enhanced prediction accuracy.