BioSyn AI: Repurposing Life

"Repurposing Life through Geometric Deep Learning."

BioSyn AI is an end-to-end generative pipeline designed to imagine novel drug candidates (ligands) that bind to specific protein targets. It combines E(n)-Equivariant Graph Neural Networks (GNNs) for protein structure encoding with 3D Denoising Diffusion Probabilistic Models (DDPMs) for molecule generation.

---

🧬 Architecture

The pipeline follows a closed-loop generative process:

graph LR
    A[Protein PDB] -->|Ingestion Engine| B(Geometric Graph)
    B -->|GNN Encoder| C[Context Embedding]
    D[Gaussian Noise] -->|Diffusion Model| E{Reverse Process}
    C --> E
    E -->|Denoising| F[3D Atom Cloud]
    F -->|KNN Builder| G[SMILES Candidate]

Loading

1. Ingestion: TypeScript engine fetches raw PDB/SDF files from biological databases.
2. Encoder: A GNN extracts geometric features (invariant to rotation/translation) from the protein pocket.
3. Decoder: A Diffusion model iteratively refines random noise into stable 3D molecular structures conditioned on the protein embedding.
4. Inference: A robust MoleculeBuilder reconstructs valid chemical graphs from 3D point clouds using K-Nearest Neighbors (KNN) logic.

---

⚡ Quick Start

Prerequisites

• Python 3.10+
• Node.js (v16+)
• CUDA-enabled GPU (Recommended)

1. Installation

Clone the repository and set up the hybrid environment.

# Clone the repo
git clone [https://github.com/zumermalik/BioSyn-AI-Repurposing-Life.git](https://github.com/zumermalik/BioSyn-AI-Repurposing-Life.git)
cd BioSyn-AI-Repurposing-Life

# Set up Python Environment (Conda recommended for RDKit compatibility)
conda create -n biosyn python=3.10 -y
conda activate biosyn

# Install Core Dependencies
pip install -r requirements.txt

# Install Ingestion Engine (TypeScript)
npm install

2. Run the Pipeline (Zero to Hero)

You can run the entire inference stack with a single command. This will load the pre-trained checkpoint and generate candidates for the target protein 5R82.

# Run Inference
python src/pipeline/inference_pipeline.py

Expected Output:

🧪 Starting BioSyn Inference on cuda...
   >> Target Protein: 5R82.pdb
   >> Loading checkpoint: checkpoints/biosyn_epoch_5.pt
   >> Generating 5 drug candidates...
      🔹 Candidate 1: CC(=O)Nc1ccc(O)cc1
      🔹 Candidate 2: CN1C=NC2=C1C(=O)N(C(=O)N2C)C
✅ Generation Complete. 5 candidates saved to results/

---

📂 Project Structure

BioSyn-AI-Repurposing-Life/
├── configs/              # Hyperparameter Configuration
├── data/                 # Data Storage
│   ├── external/         # External Databases (PDBBind/CrossDocked)
│   ├── processed/        # PyTorch Geometric Tensors
│   └── raw/              # Original PDB/SDF Files
├── notebooks/            # Jupyter Prototyping Environments
├── src/                  # Source Code
│   ├── chemistry/        # RDKit Logic & Molecule Builders
│   ├── ingestion/        # TypeScript/Python Data Fetchers
│   ├── models/           # GNN Encoder & Diffusion Decoder
│   ├── pipeline/         # Training & Inference Orchestration
│   ├── utils/            # Utility Functions
│   ├── __init__.py       # Package Initialization
│   └── main.py           # Main Application Entry Point
├── tests/                # Unit Tests
├── environment.yml       # Conda Environment Definition
├── LICENSE               # Apache 2.0 License
├── package.json          # Node.js Dependencies
├── pyproject.toml        # Python Packaging Configuration
├── README.md             # Project Documentation
├── requirements.txt      # Python Dependencies
├── ROADMAP.md            # Future Development Plans
└── tsconfig.json         # TypeScript Configuration

🛠️ Development & Testing

We use pytest for unit testing the geometric logic and chemical validity.

# Run the full test suite
pytest tests/

---

🤝 Contributing

Contributions are what make the open-source community such an amazing place to learn, inspire, and create. Any contributions you make are greatly appreciated.

1. Fork the Project
2. Create your Feature Branch (git checkout -b feature/AmazingFeature)
3. Commit your Changes (git commit -m 'Add some AmazingFeature')
4. Push to the Branch (git push origin feature/AmazingFeature)
5. Open a Pull Request

Contribution Standards

• Code Style: Please use black for Python formatting.
• Testing: Ensure all new modules have accompanying tests in tests/.
• Data: Do not commit large datasets (PDB/SDF files) to Git. Use the data/ folder.

---

📜 Citation & License

This project is licensed under the Apache 2.0 License. If you use this architecture in your research, please link back to this repository.

---

Maintained by the Builders.