Cancer Subtype Classifier using Gene Expression

Cancer Subtype Classifier – Demo using Gene Expression and ML

🧩 Problem Statement

In cancer genomics, identifying the cancer subtype of a tumor sample is crucial for:

• Accurate diagnosis
• Treatment planning
• Precision medicine research

However, high-dimensional RNA-seq data makes subtype classification challenging, especially across multiple cancer types.

---

❓ Why Are We Solving This?

Automatically classifying tumor samples based on gene expression can:

• Help researchers label or validate samples
• Support oncologists with second-opinion diagnostics
• Serve as a preprocessing layer for further mutation or survival analysis

---

Approach

We built a multi-class machine learning classifier that:

1. Takes a tumor sample’s RNA-seq gene expression (20,531 genes)
2. Predicts the primary cancer type (subtype)
3. Visualizes performance, top genes, and live predictions

---

Dataset

We used the TCGA-PANCAN-HiSeq-801x20531 dataset:

• Samples: 801 tumor samples
• Genes: 20,531 (RNA-Seq HiSeq platform)
• Source: The Cancer Genome Atlas (TCGA)
• Format: CSV matrix (samples × genes) with labels

Cancer types in this pancancer set include:

• BRCA – Breast invasive carcinoma
• COAD – Colon adenocarcinoma
• LUAD – Lung adenocarcinoma
• KIRC – Kidney renal clear cell carcinoma
• HNSC – Head and neck squamous cell carcinoma
• OV – Ovarian serous cystadenocarcinoma
• THCA – Thyroid carcinoma
  …and others

---

Tech Stack

Component	Tool
Machine Learning Model	XGBoost Classifier
Label Encoding	scikit-learn LabelEncoder
Pipeline	scikit-learn Pipeline
Dashboard UI	Streamlit
Data Processing	pandas, seaborn, matplotlib
Model Persistence	joblib

---

Working

1. Load data_small.csv (trimmed gene matrix) and labels.csv
2. Encode cancer subtypes as numeric labels
3. Train an XGBoost classifier in a Scikit-learn pipeline
4. Save the trained pipeline (classifier_pipeline.joblib)
5. Load the model in a Streamlit dashboard:
   • View performance metrics
   • See top gene importances
   • Select a sample and get predicted cancer type

---

Output

• ✅ Accuracy: ~85% on test data (varies by split)
• 🔬 Top genes shown by feature importance
• 🧠 Interactive Streamlit web app for exploration

---