Molecular Kinetics: Sampling, Design and Machine Learning
- Frank Noe (Free University of Berlin, Germany)
- Carsten Hartmann (BTU Cottbus-Senftenberg, Germany)
- Christof Schuette (FU Berlin AND Zuse Institute Berlin, Germany)
- John Chodera (Memorial Sloan Kettering Cancer Center, USA)
- Cecilia Clementi (Rice University, Houston, USA)
- Bettina Keller (Free University of Berlin, Germany)
- Gerhard Hummer (Max Planck Institute of Biophysics, Germany)
- Simon Olsson (Freie Universität Berlin, Germany)
NOTE: Registration can only be considered through the Workshop homepage linked below NOT via email.
Please visit the official website of the Workshop here:
The last few years have seen breakthroughs in our ability to simulate and measure the kinetics and functional mechanism of macromolecular systems such as proteins. Using special-purpose hardware or GPU clusters, direct molecular dynamics (MD) simulations can reach milliseconds [1-2]. With methods based on Markov modeling, transition path sampling or hyperdynamics, kinetics and mechanisms can be simulated that involve events on the timescales of seconds and beyond [3-4]. Recently, deep machine learning (ML) methods have had an impact in terms of increasing automation and accuracy of these these approaches [6-7]. Arguably, the main current limitation of the field of molecular kinetics theory is that it doesn't go beyond describing a given simulation setup, i.e. it doesn't yet attempt to scale, e.g. across the space of protein sequences which would be relevant to connect to the genomic revolution and to perform in silico design of functional properties such as enzyme turnover.
On the other hand, ML has had significant impact in the field of quantum mechanics (QM) and material design [8-9] where predicting quantities such as formation energies across chemical space is a fundamental part of the problem setting and has resulted into very sophisticated ways of learning transferable predictors that yield high accuracy and now allow QM-accuracy MD simulations [10-11]. Arguably, the main current limitation of this field is that it aims at mainly predicting microscopic quantities such as potential energies or forces. The efficient and accurate computation of thermodynamic or kinetic expectation values is likely "the next big problem" faced by this community, which would benefit from input by Statistical Mechanics and MD experts.
These two communities can learn and benefit from each other in order to develop new areas such as "transferable molecular kinetics" or "chemical quantum kinetics" that may facilitate the design of efficient enzymes, long-residence drugs or materials with properties that are not only a function of the crystal structure alone.
 K Lindorff-Larsen et al.: How fast-folding proteins fold. Science 334 (6055), 517 (2011)
 TJ Lane et al.: To milliseconds and beyond: challenges in the simulation of protein folding. Curr. Opin. Struct. Biol. 23 (1), 58-65 (2013)
 N Plattner et al.: Complete protein-protein association kinetics in atomic detail revealed by molecular dynamics simulations and Markov modelling. Nat. Chem. 9 (10), 1005 (2017)
 R Casasnovas et al.: Unbinding kinetics of a p38 MAP kinase type II inhibitor from metadynamics simulations. JACS 139, 4780-4788 (2017)
 A Mardt et al.: VAMPnets for deep learning of molecular kinetics. Nat. Communications 9, 5 (2018)
 JML Ribeiro et al.: Reweighted autoencoded variational Bayes for enhanced sampling (RAVE). J. Chem. Phys. 149 072301 (2018)
 J Behler and M Parrinello: Generalized Neural-Network Representation of High-Dimensional Potential-Energy Surfaces. Phys. Rev. Lett. 98, 146401 (2007)
 M Rupp et al.: Fast and accurate modeling of molecular atomization energies with machine learning. Phys. Rev. Lett. 108, 058301 (2012)
 KT Schütt et al.: SchNet-A deep learning architecture for molecules and materials. J. Chem. Phys. 148, 241722 (2018)
 JS Smith et al.: Outsmarting Quantum Chemistry Through Transfer Learning. ChemRXiv (2018)