materialyzeai/matgl

Graph deep learning library for materials

What it solves

MatGL provides a specialized graph deep learning framework for materials science. It addresses the need for efficient surrogate models that can predict materials properties (such as formation energy and band gaps) and simulate potential energy surfaces (PES) for atoms, which are naturally represented as mathematical graphs.

How it works

The library implements several graph neural network (GNN) architectures tailored for atomic structures, including:

  • QET: A charge-aware, equivariant architecture for systems involving charge transfer.
  • TensorNet: An O(3)-equivariant message-passing network using Cartesian tensor representations.
  • CHGNet: A network that captures atom-bond and bond-angular relations to predict atomic magnetic moments.
  • M3GNet: An invariant network incorporating 3-body interactions, capable of acting as a foundation potential across the periodic table.
  • MEGNet: A general graph network for predicting properties in molecules and crystals.

It utilizes PyTorch Geometric (PyG) as its primary backend and offers optional acceleration via NVIDIA Warp (for training and inference on GPUs) and JAX (for portable inference across CPU, CUDA, and Apple Silicon).

Who it’s for

It is designed for materials scientists and researchers who need to develop, share, or use deep learning models to predict material properties or perform molecular dynamics (MD) and structure relaxations.

Highlights

  • Diverse Architectures: Supports multiple state-of-the-art GNNs including TensorNet, M3GNet, and QET.
  • High-Performance Backends: Optional acceleration via NVIDIA Warp and JAX for significant speedups in MD and relaxation loops.
  • LAMMPS Integration: Includes a native pair_style matgl to evaluate energies, forces, and virials within the LAMMPS simulator.
  • Hugging Face Integration: Pre-trained models can be easily loaded from and published to the Hugging Face Hub.
  • CLI Tooling: Provides a command-line interface for quick property predictions and structure relaxations.

Related

  • Project
    atomistic-machine-learning/schnetpack

    A deep learning toolbox for predicting quantum-chemical properties of molecules and materials using atomistic neural networks.

  • Project
    microsoft/mattersim

    A deep learning atomistic model for simulating bulk materials across various elements, temperatures, and pressures to predict energy, forces, and stress.

  • Project
    materialyzeai/maml

    A Python package that provides high-level interfaces for materials science machine learning, enabling the conversion of materials data into features for property prediction and energy surface modeling.

  • Project
    torchmd/torchmd-net

    A PyTorch-based library providing state-of-the-art neural network potentials for molecular simulations and the tools to train them.

  • Dispatch
    Introduction to Graph Machine Learning

    Hugging Face provides a comprehensive overview of Graph Machine Learning, detailing how graphs are represented and the evolution from pre-neural features to Graph Neural Networks and Graph Transformers.