A New Perspective on Building Efficient and Expressive 3D Equivariant Graph Neural Networks

This is the official implementation of the LEFTNet method proposed in the following paper.

Weitao Du, Yuanqi Du, Limei Wang, Dieqiao Feng, Guifeng Wang, Shuiwang Ji, Carla Gomes, Zhi-Ming Ma. "A New Perspective on Building Efficient and Expressive 3D Equivariant Graph Neural Networks". [NeurIPS 2023]

Local Hierarchy of 3D Isomorphism

From Local to Global

LEFTNet implementation (LSE+FTE)

Requirements

We include key dependencies below. The versions we used are in the parentheses.

• PyTorch (1.9.0)
• PyTorch Geometric (1.7.2)

Run

QM9

device=0
target='homo' # 'mu', 'alpha', 'homo', 'lumo', 'gap', 'r2', 'zpve','U0', 'U', 'H', 'G', 'Cv'
python main_qm9.py --device $device --target $target

MD17

device=0
name='aspirin' #aspirin, benzene2017, ethanol, malonaldehyde, naphthalene, salicylic, toluene, uracil
python main_md17.py --device $device --name $name

Citation

@article{du2024new,
  title={A new perspective on building efficient and expressive 3D equivariant graph neural networks},
  author={Du, Yuanqi and Wang, Limei and Feng, Dieqiao and Wang, Guifeng and Ji, Shuiwang and Gomes, Carla P and Ma, Zhi-Ming and others},
  journal={Advances in Neural Information Processing Systems},
  volume={36},
  year={2023}
}

Acknowledgments

We acknowledge DIG library for adapting the training pipeline on QM9 and MD17.