JAMUN: Bridging Smoothed Molecular Dynamics and Score-Based Learning for Conformational Ensembles

JAMUN is a novel approach for generating conformational ensembles of protein structures, presented in the paper JAMUN: Bridging Smoothed Molecular Dynamics and Score-Based Learning for Conformational Ensembles.

Conformational ensembles of protein structures are immensely important both for understanding protein function and drug discovery in novel modalities such as cryptic pockets. JAMUN performs molecular dynamics (MD) in a smoothed, noised space of all-atom 3D conformations of molecules by utilizing the framework of walk-jump sampling. It enables ensemble generation for small peptides at rates of an order of magnitude faster than traditional molecular dynamics. The physical priors in JAMUN enable transferability to systems outside of its training data, even to peptides that are longer than those originally trained on.

Overview of walk-jump sampling in JAMUN

For the official code, setup instructions, and further details, please refer to the GitHub repository.

Usage

To use the JAMUN model for sampling conformations, you first need to set up the environment and download the pre-trained checkpoints as described in the official GitHub repository and the model collection on Hugging Face.

Once set up and checkpoints are downloaded, you can run sampling for a peptide sequence (e.g., "AGPF") as follows:

Load Trained Models: Ensure you have git-lfs installed, then clone the model collection:
```
git lfs install
git clone https://huggingface.co/ameya98/JAMUN
```
Generate a .pdb file for your peptide (if you don't have one):
```
python scripts/prepare_pdb.py AGPF --mode uncapped --outputdir .
```
This command will create a PDB file (e.g., uncapped-AGPF.pdb) in the specified output directory.
Run the sampling command: Ensure your jamun conda environment is active and necessary environment variables are set. Replace <INIT_PDB_PATH> with the path to your generated .pdb file and <CHECKPOINT_DIR> with the path to your downloaded model checkpoint (e.g., ./JAMUN/2AA_uncapped_large_model).
```
jamun_sample --config-dir=configs experiment=sample_custom \
             ++init_pdb=<INIT_PDB_PATH> \
             ++checkpoint_dir=<CHECKPOINT_DIR>
```

For more detailed information on training, inference, and analysis, please visit the official GitHub repository.

Citation

If you find this work useful, please cite the paper:

@misc{daigavane2024jamuntransferablemolecularconformational,
      title={JAMUN: Bridging Smoothed Molecular Dynamics and Score-Based Learning for Conformational Ensembles},
      author={Ameya Daigavane and Bodhi P. Vani and Darcy Davidson and Saeed Saremi and Joshua Rackers and Joseph Kleinhenz},
      year={2024},
      eprint={2410.14621},
      archivePrefix={arXiv},
      primaryClass={physics.bio-ph},
      url={https://arxiv.org/abs/2410.14621},
}