AUTALASSO

AUTALASSO applies an automatic adaptive LASSO approach for genome-wide prediction to select sparse marker effects and deliver accurate predictions of breeding values and phenotypes in animal and plant breeding.

Key Features:

• Adaptive LASSO Optimization: Builds on the adaptive LASSO framework to perform variable selection tailored to sparse genomic problems.
• ADMM Algorithm Integration: Solves the optimization problem using the alternating direction method of multipliers (ADMM) for large-scale efficiency.
• Automatic Hyper-Parameter Tuning: Automatically tunes the learning rate via a line search and optimizes the regularization factor using the Golden section search.
• Sparse Data Handling: Addresses sparsity arising from a large number of markers relative to a smaller number of evaluated individuals.
• Superior Prediction Accuracy: Demonstrates improved prediction accuracy over adaptive LASSO, standard LASSO, and ridge regression (glmnet) on simulated and real bull data.
• Genetic Gain Optimization: Supports applications aiming to increase genetic gain by providing higher prediction accuracy for selection decisions.
• GWAS Capability: Extends to genome-wide association studies (GWAS) enabling analysis of additive and dominance effects with reduced prediction error versus ordinary LASSO.

Scientific Applications:

• Genomic Selection: Predicts breeding values and phenotypes from large-scale genomic datasets to inform selection in animal and plant breeding.
• Genome-wide Association Studies (GWAS): Facilitates detection and estimation of additive and dominance marker effects with improved prediction accuracy.
• Breeding Program Optimization: Supports strategies to maximize genetic improvement by providing accurate genomic predictions for selection decisions.

Methodology:

Implements an adaptive LASSO solved with ADMM, with automatic line-search tuning of the learning rate and Golden section search optimization of the regularization factor.