Home

MolGraph is a light-weight Python package for applying graph neural networks (GNNs) on molecular graphs. It is built with, and aims to be highly compatible with, TensorFlow and Keras.

As MolGraph integrates well with TensorFlow and Keras APIs, it allows for easy, flexible and efficient implementations of GNNs, targeting a wide audience. Furthermore, a chemistry module is implemented to generate highly customizable molecular graphs.

Note

Currently, Keras 3 does not support extension types. As soon as it does, it is hoped that MolGraph will migrate to Keras 3.

Overview of MolGraph:

Get started

• Installation
• Walk through

API

• Graph Tensor
• Layers
• Models
• Chemistry

Tutorials

• Benchmark GAT model on the ESOL dataset
• Benchmark DTNN model on the QM7 dataset
• Benchmark MPNN model on the Tox21 dataset (with Masked Loss)