RFGeneRank

RFGeneRank harmonizes bulk RNA-seq matrices, optionally applies batch correction, and trains cross-validated classification models to enable stable predictive gene ranking for transcriptomics.

Key Features:

• Batch Correction: Optionally applies batch correction to harmonized bulk RNA-seq matrices.
• Cross-validated Classification: Trains classification models using ranger, glmnet, or xgboost with cross-validation.
• Leakage-safe Feature Selection: Implements feature selection that prevents data leakage across cross-validation folds.
• Permutation Importance: Computes permutation importance for feature evaluation.
• SHAP-based Interpretability: Provides SHAP-based model interpretability and SHAP dependence plots.
• Calibration Methods: Supports probability calibration using Platt scaling or isotonic regression.
• Stability Metrics: Calculates stability metrics across cross-validation folds.
• Embeddings: Generates PCA and UMAP embeddings for visualization.
• ROC Visualization: Produces ROC curve visualizations for model performance assessment.
• Ranked-gene Tables: Outputs tidy ranked-gene tables for downstream analysis.

Scientific Applications:

• Bulk RNA-seq Data Harmonization: Integrates and standardizes bulk RNA-seq expression matrices from multiple sources or batches.
• Predictive Gene Ranking: Identifies and ranks genes with stable predictive power for phenotype classification in transcriptomic studies.
• Biomarker Discovery: Supports discovery of robust gene signatures through cross-validated feature selection and stability assessment.
• Model Interpretation: Enables mechanistic interpretation of transcriptomic classifiers via SHAP analysis and dependence plots.

Methodology:

Cross-validated classification using ranger, glmnet, or xgboost with leakage-safe feature selection, permutation importance, SHAP-based interpretability, and Platt or isotonic calibration.