hgv_lps

hgv_lps applies the LASSO-Patternsearch algorithm to analyze high-dimensional genetic data from single-nucleotide polymorphisms (SNPs) and covariates, including next-generation DNA sequencing datasets, to identify informative predictors and interactions for association and classification analyses.

Key Features:

• LASSO-type penalized likelihood core: Employs a LASSO-type penalized likelihood method to manage datasets with numerous SNPs and covariates and to perform variable and interaction selection.
• Screening step (parametric logistic regression): Implements an initial screen using parametric logistic regression to filter candidate patterns.
• Penalized logistic regression (LASSO penalty): Refines screened patterns via penalized logistic regression with an LASSO penalty for sparse pattern selection.
• Final model construction (parametric logistic regression): Builds a final parametric logistic regression model from patterns that survive the LASSO step.
• Application to GAW15 Problem 3 simulated rheumatoid arthritis dataset: Applied to the Genetic Analysis Workshop 15 Problem 3 simulated rheumatoid arthritis data (100 replicates of SNP and covariate data for 1500 families with an affected sib pair and 2000 controls) and identified most associated SNPs and relevant covariates.
• Competitive classification performance: Models generated by the algorithm achieved competitive error rates when used as classifiers.
• Interaction detection: Detects interactions between predictor variables (SNPs and covariates) relevant to complex-trait analyses.
• Scalability to NGS-scale data: Designed to handle large-scale datasets, including those generated by next-generation DNA sequencing technologies.

Scientific Applications:

• Complex-trait and disease association studies: Identification of genetic markers and interactions for diseases such as rheumatoid arthritis.
• Family-based and case-control analyses: Analysis of datasets with family structures (affected sib pairs) alongside control samples for association mapping.
• Variable selection and interaction discovery in high-dimensional SNP datasets: Selection of informative SNPs and covariates and discovery of their interactions in high-dimensional settings.
• Classifier development for case-control discrimination: Construction of logistic-regression-based classifiers from selected patterns to differentiate cases and controls.

Methodology:

Explicitly performs a screening step using parametric logistic regression, applies penalized logistic regression with a LASSO penalty for pattern selection, and constructs a final parametric logistic regression model from patterns that survive the LASSO-type penalized likelihood procedure.