Skip to content

DeepDriveWE-Academy

Weighted ensemble molecular dynamics using the Academy multi-agent framework.

DeepDriveWE-Academy is a Python implementation of DeepDriveWE built on the Academy multi-agent framework. It enables scalable weighted ensemble (WE) simulations where molecular dynamics walkers and resampling logic run as independent, communicating agents distributed across CPUs and GPUs.

Key Features

  • Agent-based architecture -- Simulation walkers and the WE orchestrator are Academy agents that communicate asynchronously via typed messages (SimResult, SimMetadata).
  • Pluggable components -- Swap binners, recyclers, resamplers, and simulation engines independently.
  • GPU-distributed MD -- Simulations are offloaded to GPU workers through Parsl, with built-in support for workstations, Slurm, and PBS clusters.
  • Checkpointing and resume -- Ensemble state is saved after every iteration in both JSON and WESTPA-compatible HDF5 format.
  • AI-enhanced resampling -- Optional deep learning models (convolutional VAE, adversarial autoencoders) for latent-space--guided resampling.

Quick Example

from deepdrivewe.workflows import run_westpa_workflow

await run_westpa_workflow(
    manager=manager,
    sim_agent_type=MySimAgent,
    westpa_agent_type=MyWestpaAgent,
    max_iterations=100,
    ensemble=ensemble,
)

See the Quickstart guide to get running in minutes, or follow the full OpenMM NTL9 tutorial for a production protein-folding workflow.

Build with Claude Code

This project ships a Claude Code skill that walks Claude through creating a new DeepDriveWE example end-to-end. See the Claude Skill page to install it, then ask Claude things like "build a new DeepDriveWE example for ligand unbinding with AMBER".

Citation

If you use DeepDriveWE in your research, please cite:

Leung, J. M. G.; Frazee, N. C.; Brace, A.; Bogetti, A. T.; Ramanathan, A.; Chong, L. T. "Unsupervised Learning of Progress Coordinates during Weighted Ensemble Simulations: Application to NTL9 Protein Folding." Journal of Chemical Theory and Computation 2025, 21 (7), 3691--3699. DOI: 10.1021/acs.jctc.4c01136

BibTeX:

@article{leung2025unsupervised,
  title={Unsupervised Learning of Progress Coordinates during Weighted Ensemble Simulations: Application to NTL9 Protein Folding},
  author={Leung, Jeremy MG and Frazee, Nicolas C and Brace, Alexander and Bogetti, Anthony T and Ramanathan, Arvind and Chong, Lillian T},
  journal={Journal of chemical theory and computation},
  volume={21},
  number={7},
  pages={3691--3699},
  year={2025},
  publisher={ACS Publications}
}

License

DeepDriveWE-Academy is released under the MIT License.