OXtal-CSP/OXtal

OXtal (Organic X "Crys-" tal) is an all-atom diffusion model for molecular crystal structure prediction (CSP). Unlike traditional quantum-chemical approaches, which rely on expensive energy oracles, OXtal generates fast and accurate zero-shot predictions at a fraction of the cost. Specifically, OXtal recovers experimental crystal structures for both rigid and flexible molecules, as well as co-crystals, with conformer RMSD1 < 0.5 Å and attains over 80% packing similarity rate, demonstrating its ability to model both thermodynamic and kinetic regularities of molecular crystallization.

The model was introduced in the paper, OXtal: An All-Atom Diffusion Model for Organic Crystal Structure Prediction (ICLR 2026).

🚀 Sample Usage

To run inference, please follow the installation instructions in the official GitHub repository. You can then run generation using the provided runner:

# Run OXtal inference:
bash run_inference.sh

📖 Citation

If you make use of this code or its accompanying paper, please cite this work as follows:

@inproceedings{jin2025oxtal,
  title={OXtal: An All-Atom Diffusion Model for Organic Crystal Structure Prediction},
  author={Jin, Emily and Nica, Andrei Cristian and Galkin, Mikhail and Rector-Brooks, Jarrid and Lee, Kin Long Kelvin and Miret, Santiago and Arnold, Frances H and Bronstein, Michael and Bose, Avishek Joey and Tong, Alexander and Liu, Cheng-Hao},
  booktitle={ICLR},
  year={2026}
}

📄 License

The source code for OXtal is released under an MIT License. However, since OXtal was trained on data from CCDC's Cambridge Structural Database, the model weights are released under a [Creative Commons Attribution-NonCommercial 4.0 International][cc-by-nc] (CC BY-NC 4.0) License. For commercial use of the model weights, please ensure that you have a proper CCDC License.