HyperLasso

HyperLasso performs simultaneous analysis of multiple single nucleotide polymorphisms (SNPs) and covariates in genetic association studies using penalized regression to select predictive variants and enforce model sparsity.

Key Features:

• Multi-SNP and covariate analysis: Performs simultaneous analysis of multiple single nucleotide polymorphisms (SNPs) and covariates in genetic association studies.
• Penalized regression methods: Implements elastic net, ridge regression, Lasso (Least Absolute Shrinkage and Selection Operator), MCP (Minimax Concave Penalty), and a normal-exponential-γ shrinkage prior for variable selection and shrinkage.
• Coefficient shrinkage and sparsity: Shrinks coefficients of non-influential markers to zero to produce parsimonious models.
• Penalty tuning and model selection: Allows penalty strength selection via cross-validation or maximum likelihood-based model selection criteria such as the Bayesian Information Criterion (BIC).
• Type I error control: Includes mechanisms to control type I error during model selection.
• Simulation-based evaluation: Evaluates method performance by simulating data under varying disease locus effect sizes and linkage disequilibrium patterns.
• Handling correlated variables: Manages correlated predictors and linkage disequilibrium to avoid simultaneous entry of correlated variables, maintaining interpretability.
• Comparative benchmarking: Compares performance against standard single-locus analysis and forward stepwise regression.

Scientific Applications:

• Genetic association studies: Identifying SNPs and covariates associated with quantitative or binary traits in genetic association analyses.
• GWAS variable selection: Variable selection in genome-wide association studies with attention to controlling false positives (type I error).
• Method benchmarking: Benchmarking and comparing penalized regression approaches under different disease locus effect sizes and linkage disequilibrium patterns.
• Modeling correlated predictors: Modeling and interpreting correlated predictors arising from linkage disequilibrium to produce sparse, interpretable models.

Methodology:

Implements penalized regression approaches (elastic net, ridge regression, Lasso, MCP, normal-exponential-γ shrinkage prior), tunes penalty strength via cross-validation or maximum likelihood-based BIC, evaluates performance using simulations across disease locus effect sizes and linkage disequilibrium patterns, and assesses type I error and comparisons to single-locus and forward stepwise regression.