PocketVina

GPU-accelerated protein-ligand docking with automated pocket detection, exploring through multi-pocket conditioning.

PocketVina finds the pockets of each protein with P2Rank and docks every ligand into the best pockets on the GPU.

Installation

pip install pocketvina

PocketVina runs on Linux with an NVIDIA GPU and needs Java 17+ for P2Rank. Other ways to install (conda, Docker) are in the documentation.

Citations

If you use PocketVina, please cite our preprint:

@misc{sarigun2025pocketvinaenablesscalablehighly,
      title={PocketVina Enables Scalable and Highly Accurate Physically Valid Docking through Multi-Pocket Conditioning},
      author={Ahmet Sarigun and Bora Uyar and Vedran Franke and Altuna Akalin},
      year={2025},
      eprint={2506.20043},
      archivePrefix={arXiv},
      primaryClass={q-bio.QM},
      url={https://arxiv.org/abs/2506.20043},
}

PocketVina builds on P2Rank, QuickVina2-GPU, AutoDock Vina, PoseBusters, Mol* and RDKit. Please cite them too: their papers are on the citing page.

Getting started

• Quick start: dock the example proteins and ligands, then look at the results in 3D
• Google Colab notebook
• Command line
• Large screens on a cluster (SLURM)

Benchmarks

The benchmark data and results are on Zenodo. The code to reproduce them is in benchmark/.

Documentation

The documentation was generated with Sphinx from the Markdown pages in docs/, using MyST-Parser and the Furo theme. To build it locally:

pip install -r docs/requirements.txt
sphinx-build docs docs/_build/html

Contact

For questions and bug reports, open an issue or start a discussion.

License

PocketVina is released under the MIT License (LICENSE). The docking engine is based on AutoDock Vina and QuickVina2-GPU-2.1 and keeps their Apache License 2.0. The bundled Mol* viewer is MIT licensed.