Extended Bayesian LASSO for multiple quantitative trait loci mapping and unobserved phenotype prediction.

Area of Science:

• Genetics
• Statistical Genomics
• Bioinformatics

Background:

• The Bayesian LASSO (BL) is effective for sparse modeling in quantitative trait loci (QTL) mapping and genomic breeding value (GBV) estimation.
• A limitation of BL is its single regularization parameter controlling both model sparsity and individual effect shrinkage, which is problematic with varying predictor effect sizes.

Purpose of the Study:

• To introduce the extended Bayesian LASSO (EBL) as a more robust method for QTL mapping and phenotype prediction.
• To enhance the hierarchical specification of BL by separating model sparsity and parameter shrinkage.

Main Methods:

• Developed the extended Bayesian LASSO (EBL) with an additional hierarchical level.
• Compared EBL performance against the Bayesian LASSO (BL) and Bayesian adaptive LASSO (BAL) using simulations.

Main Results:

• EBL demonstrated superior accuracy in effect size estimates and phenotypic value predictions compared to BL.
• EBL showed improved robustness to tuning compared to BAL, which lacks a mechanism to distinguish sparsity from shrinkage.
• Computational time for EBL was comparable to BL.

Conclusions:

• EBL offers a "doubly adaptive" approach, outperforming BL and BAL in accuracy and tuning robustness for QTL mapping and GBV estimation.
• EBL represents a promising advancement for genetic analysis, phenotype prediction, and breeding value estimation.