AML & ALL Classification via Gene Expression

Machine Learning Mini-Project

 

---

---

Overview

Replication of Golub et al. (1999) — one of the most cited papers in computational biology — which demonstrated that gene expression profiling alone could classify leukaemia subtypes. This project applies modern Python tools to the same dataset to classify patients as ALL or AML.

Research Question: Can gene expression profiling alone accurately classify leukaemia patients into ALL or AML subtypes, and which genes are most informative for distinguishing between them?

---

Results

PCA — Unsupervised Analysis

Classification — Random Forest & SVM (80/20 Split)

Feature Importance — Most Informative Genes

Both models achieved 80% accuracy, closely replicating the original paper's 85%.

---

Dataset

• Source: Golub et al. (1999) — Kaggle
• Samples: 72 patients, 7,071 genes (after AFFX control probe removal)
• Split: 80/20 stratified — 57 training, 15 test

---

Usage

1. Set up Kaggle API credentials (~/.kaggle/kaggle.json)
2. Install dependencies: pip install -r requirements.txt
3. Run: jupyter notebook project.ipynb

---

References

• Golub, T.R. et al. (1999). Molecular Classification of Cancer. Science, 286(5439), 531–537.
• Crawford, J. (2017). Gene Expression Dataset. Kaggle.