About the project

torch-molecule brings molecular property prediction, representation learning, and generation into a shared Python interface. It combines PyTorch models with SMILES inputs, training utilities, dataset loaders, and checkpoint persistence.

The project is developed openly on GitHub. Browse the Choose a model for supported workflows or read the class references for method papers and original implementations.

Contribute

Report an issue with a small reproducible example, your package version, and the relevant error output. Model integrations, documentation improvements, and additional examples are welcome through pull requests.

For a local development installation:

micromamba activate torch-molecule
python -m pip install -e ".[dev]"

Keep method-specific code with its model, follow comparable classes, and add focused tests and a usable example. See Installation for building the docs.

Contributors

The following contributors have added model families and data utilities:

  • Man Hei Matthew Thom — pretrained generators and dataset splitting.

  • Eric Inae — Moama, GraphMAE, AttrMask, ContextPred, and EdgePred.

  • Yihan Zhu — DeFoG, GRIN, BFGNN, RPGNN, and the SMILES Transformer.

See the repository contributors for the current contributor list.

License and acknowledgements

The project is distributed under the MIT license. Integrated methods retain their accompanying licenses and attribution. The initial project structure was adapted from python-project-template.