GTC 2020: Hybrid Molecular Mechanics: Artificial Intelligence Simulation Methods to Study Molecular Systems
After clicking “Watch Now” you will be prompted to login or join.
Click “Watch Now” to login or join the NVIDIA Developer Program.
Hybrid Molecular Mechanics: Artificial Intelligence Simulation Methods to Study Molecular Systems
Julio Maia, University Of Illinois at Urbana-Champaign | David Hardy, University of Illinois at Urbana-Champaign
The advent of modern GPU architectures, and the development of molecular dynamics (MD) engines to exploit them efficiently, substantially impacted the performance and timescale achieved by MD. We'll present recent enhancements of MD capabilities by coupling to artificial intelligence methods. We'll present the interfaces enabling the coupling of the MD engine NAMD with the deep learning-based molecular descriptor Accurate NeurAl networK engINe (ANI). By using ANI to describe unparameterized molecules in the system — a drug bound to an enzyme, for example — instead of using a more expensive method such as quantum mechanical calculations, one can achieve performances of (nearly) classical MD simulations while maintaining quantum mechanical levels of accuracy. Moreover, the new NAMD platform lets you use AI-based enhanced sampling methods, such as reinforcement learning-based adaptive sampling, to achieve timescales prohibited by pure MD simulations.