paper | blog

An LLM agent to execute protein-design tasks with RosettaScripts.

Installation

To install Agent Rosetta's agr command-line interface, run:

> uv tool install -e .
> agr setup     # Specify API keys

Hello, Rosetta!

To verify Agent Rosetta is correctly installed, run:

> agr run configs/hello-rosetta.yaml

This task will instruct Agent Rosetta to explain its capabilities:

Running Protein-design Tasks

First, setup the paths to the cxx11 and mpi version of RosettaScripts (RosettaScripts build instructions):

> agr setup rosetta

This repo contains configuration files for the tasks presented in our paper:

• configs/fixed-backbone-sequence-designs-{local/slurm}.yaml for fixed backbone sequence design with canonical amino acids only (requires a GPU to run ESMFold).
• configs/pack-ncaa-{local/slurm}.yaml for packing TRF in the core of a protein (does not require a GPU).

Both tasks use the mpi version of RosettaScripts to generate ensembles of 128 candidate designs at each step:

• The -local configs run RosettaScripts locally. Make sure the host machine has at least 130 processes available for MPI.
• The -slurm configs run RosettaScripts as SLURM jobs. Write your AGR_SLURM_PARTITION environment variable in a .env file.

Then, for example, run:

> agr run configs/pack-ncaa-slurm.yaml

Links

• Paper: arXiv:2603.15952
• Blog: Polymathic AI website

Citation

If you find this project useful, please cite:

@article{teneggi2026protein,
  title={Protein Design with Agent Rosetta: A Case Study for Specialized Scientific Agents},
  author={Teneggi, Jacopo and Turzo, SM and Marwah, Tanya and Bietti, Alberto and Renfrew, P Douglas and Mulligan, Vikram Khipple and Golkar, Siavash},
  journal={arXiv preprint arXiv:2603.15952},
  year={2026}
}

Acknowledgements

We thank Lucy Reading-Ikkanda and Aditya Chhatrala for their contribution to the artwork.

License

This project is licensed under the MIT License.