This is a implementation of our paper "Three-Dimensionally Embedded Graph Convolutional Network for interpreting spatial topology of molecules":

Hyeoncheol Cho, Insung S. Choi, Three-Dimensionally Embedded Graph Convolutional Network (3DGCN) for Molecule Interpretation

• Tensorflow
• Keras
• RDKit

• FreeSolv
• ESOL (= delaney)
• BACE
• HIV

Note that bace and bace_rotated in the code correspond to BACE_algn and BACE on paper, respectively.

See the experiment folder for training and evaluation demos of a 3DGCN model on three datasets.

If you use 3DGCN in your research, please cite:

@article{cho2018three,
  title={Three-Dimensionally Embedded Graph Convolutional Network (3DGCN) for Molecule Interpretation},
  author={Cho, Hyeoncheol and Choi, Insung},
  journal={arXiv preprint arXiv:1811.09794},
  year={2018}
}