ExhauFS

ExhauFS performs exhaustive combinatorial feature selection to identify predictive feature subsets for classification and survival analysis, reducing overfitting in machine learning models.

Key Features:

• Exhaustive Search Approach: Evaluates all possible combinations of features to identify optimal models for classification and survival tasks.
• Classification and Survival Regression Support: Supports classification tasks and Cox survival regression for survival analysis.
• Compatibility with microarray and RNA-seq: Applied to and compatible with multi-cohort microarray and RNA-seq datasets.
• Gene and isomiR Signature Construction: Enables construction of gene signatures for 5-year recurrence classification and isomiR signatures for overall survival prediction.
• Model Evaluation by Accuracy Metrics: Selects optimal feature subsets and models based on accuracy metrics.

Scientific Applications:

• Toy Cervical Cancer Dataset: Used to illustrate fundamental concepts of exhaustive feature selection.
• Breast Cancer Gene Signatures: Applied to multi-cohort microarray and RNA-seq datasets to develop gene signatures for predicting 5-year recurrence classification.
• Colorectal Cancer Survival Prediction: Utilized Cox survival regression models to construct isomiR signatures for forecasting overall patient survival.

Methodology:

Applies an exhaustive search technique that systematically evaluates all feature combinations and selects optimal models based on accuracy metrics, using Cox survival regression for survival analysis.