radical-collaboration/DeepDriveMD

DeepDriveMD in collaboration with ANL and ORNL

DeepDriveMD

DeepDriveMD, Deep learning-driven Adaptive Molecular Simulations for Protein Folding, is a toolkit developed by Brookhaven National Laboratobry (BNL)/RADICAL Laboratory at Rutgers, in collaboration with Argonne National Laboratory.

This is a main repository of DeepDriveMD used in production runs. You can find the latest publication here:

H. Lee, M. Turilli, S. Jha, D. Bhowmik, H. Ma and A. Ramanathan, "DeepDriveMD: Deep-Learning Driven Adaptive Molecular Simulations for Protein Folding," 2019 IEEE/ACM Third Workshop on Deep Learning on Supercomputers (DLS), Denver, CO, USA, 2019, pp. 12-19, doi: 10.1109/DLS49591.2019.00007. https://ieeexplore.ieee.org/abstract/document/8945122

COVID-19 Development

There is ongoing activity in associcated with the Covid-19 project. https://github.com/2019-ncovgroup/DrugWorkflows/tree/devel/workflow-2

Dependency

• OpenMM
  • swig 3+
  • numpy
  • cython
• tensorflow-gpu
  • keras
• MDAnalysis
  • scipy
  • numpy 1.16+
• scikit-learn
• Parmed
• pytables
• h5py

Systems

• fs-peptide/vhp: https://github.com/radical-collaboration/DeepDriveMD/tree/master/microscope/experiments
• ntl9: https://github.com/radical-collaboration/DeepDriveMD/tree/master/src/mdrun/ntl9

Experiment

• OLCF Summit Performance Analysis

RADICAL-Cybertools (RCT)

• EnTK documentation
• Use-case description