AUTALASSO: an automatic adaptive LASSO for genome-wide prediction

Area of Science:

• Genomics
• Animal Breeding
• Plant Breeding

Background:

• Genome-wide prediction (GWP) is crucial in breeding, utilizing large genomic datasets.
• Model sparsity is a challenge when the number of individuals is less than the number of markers.
• LASSO (Least Absolute Shrinkage and Selection Operator) is effective for sparse problems but hyper-parameter optimization is demanding.

Purpose of the Study:

• To develop a novel, computationally efficient, and automatic adaptive LASSO (AUTALASSO) method for genomic prediction.
• To optimize hyper-parameters for the ADMM algorithm used in AUTALASSO.

Main Methods:

• Developed AUTALASSO using the alternating direction method of multipliers (ADMM) optimization algorithm.
• Implemented automatic hyper-parameter tuning: learning rate with line search and regularization factor with Golden section search.

Main Results:

• AUTALASSO demonstrated superior prediction accuracy on simulated and real data compared to adaptive LASSO, LASSO, and ridge regression.
• AUTALASSO offers a flexible and computationally efficient approach to GWP.

Conclusions:

• AUTALASSO enhances prediction accuracy and genetic gain in GWP.
• AUTALASSO can perform Genome-Wide Association Studies (GWAS) for additive and dominance effects with reduced prediction error compared to standard LASSO.