Reproducible Molecular Simulations with the Help of Python
summary
Shows how Python plus OpenMM ecosystem tools (OpenMM, openmmtools, perses) can reduce the effort to parametrize and run molecular dynamics: example scripts that run a full simulation in about 60 lines and add ML-derived forces in under 20 lines. Describes training a PyTorch graph neural network to predict force-field parameters to avoid many parametrization simulations, and explains hybrid ML/classical interactions supported by utilities such as createMixSystem; cites an open pandemic pipeline (esap) that ran end-to-end without patents.
AI-written summary · about
chapters (17)
- 0:00Introduction
- 1:11Python for biomolecule simulation
- 3:40Open science in cancer research
- 5:24Biotech innovation challenges
- 7:30Open drug discovery workflow
- 9:15Molecular simulations with proteins
- 11:41Legacy biology data formats
- 13:26Python for biological data
- 15:53Efficient protein data structures
- 16:13Molecular dynamics and Newton's laws
- 18:02Force fields and standard formats
- 19:05Python to simplify simulations
- 21:18Accelerated drug discovery
- 22:24Machine learning in simulations
- 24:56Hybrid systems with ML
- 27:03Python saves lives
- 27:46Q&A
related talks
31mThe universe as balls and springs: molecular dynamics in Python
28mHow Is Python Transforming Materials Modeling with Machine Learning?
11mImproving Drug Discovery with Machine Learning...
31mScientific Machine Learning With NVIDIA PhysicsNeMo
56mWrap up and reflection talk
01h 04mNonlinear Model Reduction: Using ML to Enable Rapid Simulation of Extreme-Scale Physics Models
01h 16mReproducible machine learning and science with python
32mProbabilistic and Differentiable Programming in Scientific Simulators