Artificial Intelligence-Assisted Discovery of Molecular Mechanisms from Computer Simulations
Speaker: Dr. Roberto Covino (Max Planck Institute of Biophysics, Frankfurt, Germany)
Event details from: Artificial Intelligence-Assisted Discovery of Molecular Mechanisms from Computer Simulations
- Date
- 28 June 2019
- Time
- 11:00
Abstract
Computer simulations are a powerful instrument to discover reaction mechanisms of molecular
events. Increasing computational power and smart software enable us to simulate
longer time scales, resulting in ever larger data in the form of trajectories. Our ability to
quantitatively interpret these data, however, lags behind, because it still almost completely
relies on careful visual inspection by human experts. We need to overcome this interpretation
bottleneck to manage and take advantage of the data-avalanche expected in the Exascale
era. Obtaining a mechanistic understanding is not only the ultimate goal of many
simulations but also enables sophisticated enhanced sampling schemes, which give access
to time scales that cannot be sampled by straightforward simulations. I will introduce
a new computational framework that combines deep learning, transition path sampling
simulations, and statistical inference to accelerate the simulation of interesting molecular
events and simultaneously learn their mechanism. I will report on solutions to make the
trained neural networks interpretable by distilling simple reaction coordinates from the
simulated data. This new computational framework represents a step forward towards a
fully autonomous data driven production and interpretation of computer simulations of
molecular events.
Venue: Science Park 904, room D1.115