Improving Drug Discovery with Machine Learning...
summary
D.E. Shaw Research pairs atomistic molecular dynamics and PyTorch models to speed and improve drug design: Anton 3: 1 ms of chemical time in 6–9 days versus 10–24 years on top 2024 GPUs, ML force fields trained to quantum energies and forces with autograded negative gradients, and a subcomponent energy-decomposition scheme for labels. PyTorch Lightning, PyTorch Geometric, Kemprop, Megatron and other tools appear in pipelines. MD-augmented models raise accuracy in low-data regimes and the platform has produced four drugs now in clinical trials.
AI-written summary · about
chapters (11)
- 0:00Dshaw Research introduction
- 1:02Drug discovery challenges
- 2:07Computation in drug discovery
- 3:11Anton supercomputer impact
- 3:54Machine learning force fields for molecular dynamics
- 4:37Training with PyTorch and equivariant models
- 6:01Improving machine learning with molecular dynamics data
- 7:45MD augmentation for structure prediction
- 8:46Foundation models for drug discovery
- 9:27Clinical outcomes at Dshaw
- 9:48Q&A on MD data encoding
related talks
28mHow Is Python Transforming Materials Modeling with Machine Learning?
33mReproducible Molecular Simulations with the Help of Python
08mEngineering the Future of Pharma: AI-Driven Drug Discovery
30mMachine Learning: A New Approach to Drug Discovery
36mAlgorithmic Alchemy: Transforming Drug Discovery with Open AI
55mGPU-accelerated virtual drug screening
31mML methods to accelerate the drug discovery cycle
16mBeyond Patterns: The Case for Physics-Grounded AI in Drug Discovery