SIVS

SIVS performs stable iterative variable selection to reduce high-dimensional feature spaces in omics and clinical datasets for biomarker discovery and predictive model building.

Key Features:

• Iterative Approach: Employs an iterative variable-selection methodology that enhances stability against variability introduced during cross-validation.
• Integration of Machine Learning Techniques: Leverages machine learning algorithms with embedded feature-reduction capabilities to systematically shrink the feature space.
• Comparative Performance Assessment: Demonstrated feature sets that are on average 41% smaller than those from Least Absolute Shrinkage and Selection Operator (LASSO), Boruta, and caret Recursive Feature Elimination (RFE) without compromising model performance.
• Application to Omics and Clinical Data: Assessed across omics and clinical datasets to identify compact, informative feature subsets for downstream analysis.

Scientific Applications:

• Genomics: Selection of informative genetic features from high-dimensional genomics datasets.
• Proteomics: Reduction of proteomic feature spaces to a robust subset for biomarker and model development.
• Metabolomics: Distillation of metabolomic features to enable interpretable downstream analyses.
• Biomarker Discovery: Identification of compact candidate biomarker panels from combined omics and clinical data.
• Predictive Modeling: Construction of parsimonious predictive models with preserved performance and improved interpretability.
• Personalized Medicine: Facilitation of patient stratification and individualized predictions through stable, reduced feature sets.

Methodology:

Implements an iterative variable-selection procedure that accounts for cross-validation variability, integrates machine learning algorithms with embedded feature-reduction, and was compared against LASSO, Boruta, and caret RFE.