Tolcha-Adere/Tolykoo

ML for Bioinformatics # This repository contains the machine learning pipeline I developed as part of my MSc thesis in Bioinformatics at Addis Ababa University. The goal of this project was to identify breast cancer-associated genes from RNA-Seq data using an integrative computational framework that combines differential gene expression analysis, supervised machine learning, and systems biology. --- ## About the Project Breast cancer remains the most commonly diagnosed cancer among women worldwide, and in Ethiopia specifically, it is now the leading cause of cancer-related death among women. This study set out to identify a biologically meaningful, computationally validated panel of breast cancer-associated genes using publicly available TCGA-BRCA RNA-Seq data, with the broader goal of supporting future development of affordable, molecularly informed diagnostic tools for resource-limited healthcare settings. The workflow integrates three analytical phases: 1. **Differential Gene Expression (DGE) Analysis** — performed in Galaxy using the limma-voom pipeline 2. **Machine Learning-Based Feature Prioritization** — this repository (Random Forest, XGBoost, SVM, and SHAP interpretability) 3. **Functional and Pathway Enrichment Analysis** — GO, KEGG, Reactome, STRING, and ChEA This repository specifically covers **Phase 2: the machine learning and biomarker discovery pipeline**. --- ## Pipeline Overview | Step | What it does | |------|---------------| | 1–2 | Load TCGA-BRCA expression matrix, DEG list, and clinical metadata | | 3–4 | Filter to the 1,650 significant DEGs, transpose, and Z-score normalize | | 5–6 | Stratified 80/20 train-test split, then SMOTE applied to the training set only | | 7–9 | Train Random Forest, XGBoost, and SVM classifiers with cross-validation | | 10 | Visualize model performance (accuracy, ROC-AUC, confusion matrix, CV folds) | | 11 | Gini and SHAP feature importance analysis | | 12–13 | Three-tier biomarker funneling: 1,650 DEGs → 98 genes → 20 genes → 5 …