Replaces traditional descriptor-based machine learning models with deep learning on molecular and crystal graphs. It implements the MatErials Graph Network architecture to predict physical properties across the periodic table with high accuracy.
Tech Stack
Jupyter NotebookChemistry & Materials
Why megnet?
• State-of-the-art accuracy for crystal properties
• Handles arbitrary sized molecules/crystals
• Pre-trained models available for quick start
Limitations
• Requires TensorFlow 2.x knowledge
• Training requires significant GPU resources
• Limited support for non-periodic systems
3/6/2026
Last Update
162
Forks
21
Issues
BSD-3-Clause
License
Financial Leak Detected
Stop the "SaaS Tax"
Your team could be burning cash. Switching to megnet instantly boosts your runway.
