🧬 CAFA 6 Protein Function Prediction

Predict protein function from amino acid sequences using ESM2 protein language models and XGBoost classifiers.

🎯 Competition

CAFA 6 Protein Function Prediction on Kaggle

📊 Approach

• Embeddings: ESM2 protein language model
• Classification: Multi-label XGBoost
• Evaluation: CAFA-specific metrics (F1, Precision, Recall)

🚀 Quick Start

Installation

# Clone repository
git clone https://github.com/menna890/CAFA6-Protein-Function-Prediction.git
cd CAFA6-Protein-Function-Prediction

# Install dependencies
pip install -r requirements.txt

Download Data

# Using Kaggle API
kaggle competitions download -c cafa-6-protein-function-prediction
unzip cafa-6-protein-function-prediction.zip -d data/raw/

Run Pipeline

# Extract embeddings
python scripts/extract_embeddings.py

# Train model
python scripts/train.py

# Make predictions
python scripts/predict.py

📁 Project Structure

├── data/               # Data files (not tracked)
├── src/                # Source code
│   ├── data/          # Data loading
│   ├── features/      # Feature extraction
│   ├── models/        # Model definitions
│   └── utils/         # Utilities
├── notebooks/          # Jupyter notebooks
├── scripts/           # Executable scripts
├── models/            # Trained models (not tracked)
└── outputs/           # Results and reports

🔬 Methodology

1. Data Loading: Parse FASTA and TSV files
2. Embeddings: Extract ESM2 embeddings (480-dim vectors)
3. Classification: Multi-label Binary Relevance with XGBoost
4. Evaluation: CAFA metrics with cross-validation

📈 Results

Model	F1-Macro	F1-Micro	Precision	Recall
Baseline	TBD	TBD	TBD	TBD

🛠️ Technologies

• Python 3.9+
• PyTorch
• Transformers (HuggingFace)
• XGBoost
• scikit-learn
• BioPython

👤 Author

Menna - @menna890

📝 License

MIT License

🙏 Acknowledgments

• CAFA competition organizers
• Meta AI for ESM2
• Kaggle community