GO2Sum 2.0

GO2Sum 2.0 is a deep learning framework for generating concise, human-readable summaries of protein functions using Gene Ontology (GO) terms. This version builds upon the original GO2Sum by integrating transformer models like T5, TinyLlama, and a custom T5-MoE (Mixture of Experts) model to improve generalization and interpretability.

🚀 Features

• Summarization of GO document descriptions for protein function prediction.
• Multi-model support: T5, TinyLlama, and MoE-enhanced variants.
• Clean, modular training scripts for each architecture.
• Preprocessing utilities for GO term documents.
• Support for ROUGE-based evaluation and dataset filtering.

📁 Project Structure

GO2Sum_2.0/
├── data/                   # Training, validation data
├── preprocessing/          # Scripts for preprocessing inputs
├── dataset.py              # GODocDataset class for input handling
├── train_t5.py             # Standard T5 training
├── train_t5_moe.py         # T5 with Mixture of Experts
├── train_tinyllama.py      # TinyLlama training script
├── generate_go_doc.py      # Generate GO document from GO term list
├── requirements.txt        # Python dependencies
├── check_conda_versions.sh # Utility to verify conda versions
├── README.md               # This file
└── ...

📦 Installation

1. Clone the repository:

   git clone https://github.com/SwagarikaGiri/GO2Sum_2.0.git
   cd GO2Sum_2.0

2. Create and activate Conda environment:

   conda create -n go2sum_env python=3.10
   conda activate go2sum_env

3. Install required Python packages:

   pip install -r requirements.txt

🧪 Dataset Format

Required files under data/:

• train_dataset_cleaned.tsv
• val_dataset_cleaned.tsv
• gene_ontology.obo

Each row in TSV must contain:

• input — GO term list
• target — Human-readable functional summary

🏋️‍♀️ Training

• T5 model:

  python train_t5.py

• T5 with Mixture of Experts:

  python train_t5_moe.py

• TinyLlama model:

  python train_tinyllama.py

📝 Generating GO Documents

Generate textual descriptions from GO term list using:

python generate_go_doc.py

📊 Evaluation

You can compute ROUGE scores using:

python train_casual_rouge.py

🔧 Environment Check

Verify conda environment versions:

bash check_conda_versions.sh

🤝 Contributing

Contributions and feature suggestions are welcome!

1. Fork the repository
2. Create a new branch: git checkout -b feature/my-feature
3. Commit your changes
4. Push to the branch
5. Open a pull request

📄 License

This project is licensed under the MIT License.

👩‍🔬 Maintainer

Swagarika Giri
PhD Candidate, Purdue University
🔗 GitHub