PredictBC: Predictive Ensemble Model for Breast Cancer Classification Using Microbial and Metabolomic Signatures

Project Overview

PredictBC leverages advanced ensemble machine learning methods to classify breast cancer cases (pre- and postmenopausal) against control subjects using microbial species and metabolite (BGC) abundance data. The ensemble method significantly enhances predictive performance by integrating predictions from multiple classifiers.

---

Data Description

The data includes abundance matrices for microbial species and metabolites (BGC), as well as metadata separately for premenopausal and postmenopausal groups. These datasets distinguish between cases and controls:

• Premenopause_Case vs Premenopause_Control
• Postmenopause_Case vs Postmenopause_Control

---

Modeling Approach

The ensemble model incorporates Random Forest, Gradient Boosting, and XGBoost classifiers, employing a stacking ensemble approach to enhance predictive accuracy.

Steps Included in the Ensemble Modeling Script:

• Data preprocessing and feature engineering
• Implementation and training of individual classifiers (RF, GB, XGB)
• Stacking ensemble to combine classifier predictions
• Evaluation using ROC-AUC and confusion matrix
• Identification of significant predictive features

---

Results Interpretation

• Confusion Matrix: Illustrates classification performance of the stacking ensemble model.
• ROC Curves: Show performance comparison among individual classifiers (Random Forest, Gradient Boosting, XGBoost).
• Feature Importance: Highlights the most significant microbial and metabolite predictors contributing to the classification.

---

Running the Analysis

1. Clone the repository.
2. Navigate to the scripts/ directory.
3. Execute the Jupyter notebook (Real_Premen_Species_stack.ipynb) after placing your data in the data/ folder.

---

Dependencies

• Python (>=3.8)
• Pandas
• NumPy
• Scikit-learn
• XGBoost
• Matplotlib

Install required libraries using:

pip install pandas numpy scikit-learn xgboost matplotlib

---

Contribution

Contributions are welcome. Please open issues for suggestions or submit pull requests for enhancements.

---

License

This project is open-source. You're encouraged to freely use, adapt, and modify the scripts provided. Please acknowledge this repository if you use or modify its scripts for research or publication.