SnapBind 🧬

2nd place Global AI Hackathon (Venture Capital Track) by MIT Sloan AI Club and HackNation (2025)

Advanced Protein Binding Site Prediction with Deep Learning

SnapBind is a web application that predicts protein binding sites using a Fast Convolutional Neural Network (FastCNN). The application features an interactive web interface with real-time visualization, 3D protein structure rendering, and comprehensive data export capabilities.

🌟 Features

🔬 Advanced Machine Learning

• FastCNN Architecture: Optimized convolutional neural network with residual blocks
• GPU Acceleration: Supports CUDA, MPS (Apple Silicon), and CPU inference

🎨 Interactive Web Interface

• Real-time Predictions: Instant binding site analysis
• Adaptive Heatmaps: Responsive amino acid visualization that scales with sequence length
• Interactive Charts: Chart.js-powered probability plots

📊 Comprehensive Data Export

• Raw Output Access: Complete prediction data in JSON format
• One-Click Copy: Clipboard integration for easy data sharing
• File Downloads: Timestamped JSON exports for analysis
• Structured Results: Organized binding site information with confidence scores

🚀 Quick Start

Prerequisites

• Python 3.8 or higher
• PyTorch 2.0+
• Modern web browser with WebGL support

Installation

1. Clone the repository:

   git clone https://github.com/gitexa/snapbind.git
   cd snapbind

2. Create a conda environment:

   conda env create -f environment.yaml
   conda activate esm-mps

   Or install with pip:

   pip install -r requirements.txt

Running the Application

1. Start the Flask server:

   python app.py

2. Open your browser: Navigate to http://localhost:5004

3. Enter a protein sequence:

   Example: MKWVTFISLLFLFSSAYSRGVFRRDAHKSEVAHRFKDLGE
   

4. View results:

   • Interactive heatmap showing binding probabilities
   • Downloadable raw prediction data

📁 Project Structure

snapbind/
├── app.py                      # Flask web application
├── backend/
│   └── pred.py                # FastCNN model and prediction logic
├── templates/
│   └── index.html             # Web interface with interactive features
├── outputs/                   # PDB files and prediction outputs
├── checkpoints/               # Trained model weights
├── data/                      # Training datasets
├── training/                  # Model training scripts and notebooks
├── requirements.txt           # Python dependencies
├── environment.yaml           # Conda environment specification
└── README.md                  # This file

Built with ❤️ for the scientific community
_{Advancing protein research through machine learning}