stabs

stabs implements stability selection in R to perform variable selection with finite-sample error control for high-dimensional (n ≪ p) statistical models such as Lasso and boosting.

Key Features:

• Stability Selection Framework: Implements stability selection to identify influential variables in high-dimensional settings and integrates with selection methods such as Lasso and boosting.
• Finite Sample Error Control: Uses resampling procedures to provide finite-sample error control and manage error bounds for selected variables.
• Simulation Studies and Practical Insights: Includes evaluation through simulation studies that assess performance across varied scenarios.
• Parameter Sensitivity Analysis: Supports analysis of parameter effects including sample size, number of truly influential variables, and tuning parameters on stability selection outcomes.
• Application to Real-World Data: Has been applied to phenotype measurements in autism spectrum disorder using a log-linear interaction model fitted by boosting and identified five differentially expressed amino acid pathways.
• Flexibility for Linear and Additive Models: Provides per-family error rate control and implements complementary pairs stability selection, facilitating use with linear and additive models.

Scientific Applications:

• Ecological Studies: Variable selection in observational ecological studies requiring flexible, non-linear models.
• Biotechnological Research: Analysis of large-scale genomic or proteomic datasets with numerous predictors.
• Medical Data Analysis: Identification of significant biological pathways and predictors in medical studies, exemplified by the autism spectrum disorder phenotype analysis that found five differentially expressed amino acid pathways.

Methodology:

Uses stability selection and complementary pairs stability selection, resampling procedures for finite-sample error control, Lasso and boosting as underlying selection algorithms, simulation studies for evaluation, and fitting of log-linear interaction models by boosting in applied analyses.