DiffuCat is a next-generation scientific intelligence platform designed to accelerate the discovery of high-performance catalysts. By combining Graph Neural Networks (GNNs) with immersive 3D molecular visualization, DiffuCat enables researchers to predict, simulate, and optimize complex chemical topologies with unprecedented speed and precision.
| Service | Status | Link |
|---|---|---|
| Scientific Dashboard | 🟢 Live | https://diffucat.vercel.app/ |
| Inference Engine | 🟢 Operational | https://diffucat-fastapi.onrender.com |
A high-fidelity "Command Center" for molecular research.
- Structural Analysis: Real-time prediction of catalyst activity, selectivity, and stability.
- Inference Stream: Live tracking of model performance and candidate generation.
- High-Contrast Topology: Immediate access to SMILES metadata and structural descriptors.
The bridge between digital prediction and physical synthesis.
- Discovery Pipeline: Submit candidate batches for virtual or automated lab testing.
- Simulation Engine: Track chemical reaction trajectories and materialization status.
- Structural Archive: A metric-driven database for storing and filtering successful catalyst designs.
State-of-the-art molecular rendering powered by 3Dmol.js.
- Uncertainty Heatmaps: Direct projection of model confidence onto the molecular structure (Green = Confident, Red = Uncertain).
- Multi-Modal Rendering: Switch between Stick, Sphere, and Cross representations with a single click.
- Atomic Telemetry: Interactive HUD displaying real-time topology metadata and active scanning states.
- Frontend: Next.js 14/15, Tailwind CSS, Framer Motion, 3Dmol.js.
- Backend: FastAPI (Python), Graph Neural Networks (GNNs).
- Data Management: DVC (Data Version Control) for reproducible research.
- DevOps: Docker, Celery (for asynchronous prediction pipelines).
This guide is designed for everyone—from technical engineers to research scientists.
- Download the Code: Click the green "Code" button and select "Download ZIP", or use the command:
git clone https://github.com/Aspirant200715/DiffuCat.git
- Install Python: Ensure you have Python 3.9+ installed on your system.
- Install Node.js: Ensure you have Node.js 18+ installed on your system.
The backend handles the AI predictions and chemical simulations.
- Open your terminal/command prompt.
- Navigate to the project root directory.
- Create a Virtual Environment (Highly Recommended):
python -m venv .venv .venv\Scripts\activate # On Windows source .venv/bin/activate # On Mac/Linux
- Install Dependencies:
pip install -r requirements.txt
- Start the Inference Server:
The server will typically start at
python -m src.backend.main
http://localhost:8002.
The frontend provides the beautiful dashboard and 3D visualization.
- Open a new terminal window.
- Navigate to the
frontendfolder:cd frontend - Install Dependencies:
npm install
- Start the Interface:
npm run dev
- Access the Platform: Open your browser and go to
http://localhost:3000.
- Enter SMILES: Input a chemical SMILES string (e.g.,
CCN) into the Discovery Console. - Predict: Click "Analyze Structure" to trigger the GNN inference engine.
- Visualize: Use the 3D Viewer to inspect the catalyst. Look for Red Areas—these indicate where the model is uncertain and where more data might be needed.
- Simulate: Submit successful candidates to the Laboratory for batch processing.
DiffuCat is built with a focus on explainable AI in chemistry. By visualizing uncertainty directly on the atomic structure, we empower scientists to understand why a model makes a prediction, moving beyond "Black Box" AI.