tools-only Feature Selection Usage Guide
Overview
Select informative features for biomarker discovery using Boruta all-relevant selection, mRMR minimum redundancy, and LASSO regularization.
Prerequisites
pip install Boruta mrmr-selection scikit-learn pandas numpy
Quick Start
Tell your AI agent what you want to do:
- "Select biomarker genes using Boruta"
- "Find a minimal set of non-redundant features with mRMR"
- "Use LASSO to find sparse biomarkers"
- "Which features are stable across bootstrap samples?"
Example Prompts
Boruta Selection
"Run Boruta on my expression matrix to find all genes relevant for predicting disease status. I want the complete set of biomarkers, not just a minimal subset."
"Use Boruta feature selection on my data. How many genes are selected vs tentative?"
mRMR Selection
"Select 50 non-redundant biomarker genes using mRMR from my expression data."
LASSO Biomarkers
"Use LASSO with cross-validation to find a sparse set of genes that predict my outcome. What alpha was selected?"
Stability Selection
"Run stability selection with 100 bootstrap samples. Which features are selected in more than 60% of runs?"
Combined Approaches
"First filter to top 5000 genes by ANOVA, then run Boruta on the filtered set."
What the Agent Will Do
- Load expression matrix and sample labels
- Apply appropriate feature selection method
- Report number of selected features
- Rank features by importance/stability
- Save selected feature list
Tips
- Boruta finds ALL relevant features; mRMR/LASSO find minimal sets
- Pre-filter with univariate tests (top 1000-5000) before Boruta on large matrices
- mRMR is good when you want exactly K features with low redundancy
- LASSO picks arbitrarily among correlated features; stability selection helps
- Consider running multiple methods and taking intersection or union
- Validate selected features on independent data or nested CV