A04 - Efficient calculation of slow and stationary scales in molecular dynamics
Head(s): Prof. Dr. Frank Noé (FU Berlin), Dr. Thomas Weikl (MPI)
Project member(s): Moritz Hoffmann, Andreas Mardt, Katarzyna Ziolkowska
Participating institution(s): FU Berlin, MPIKG Potsdam
Molecular dynamics (MD) simulation is a technique that may aid in the understanding of fundamental processes in biology and chemistry, and has important technological applications in pharmacy, biotechnology, and nanotechnology. Molecular processes are often highly multi-scale with timescales spanning 15 orders of magnitude beyond the integration time-step with no pronounced gap. The previous funding period has focused on the efficient calculation of slow and stationary scales in MD. Key developments in past few years, including our own, have transformed this area and solved some of the fundamental sampling and modeling problems.
As a result, we consider the original goal of calculating slow and stationary scales in MD largely solved for small to medium protein systems. In the next funding period we will focus on four aspects: (1) To develop multilayer (deep) learning models of molecular kinetics that are both highly accurate and can describe the emergence of kinetics from molecular structure by employing transferable modeling units, similar as in MD force fields. (2) To learn physically interpretable, simulatable models of the coarse-grained dynamics that can make predictions for perturbed molecules such as mutants. (3) To apply our methods on extensively simulating structures, dynamics and mechanisms of reversible protein-protein association. (4) To develop approximation methods that are applicable beyond equilibrium dynamics and employ them to study the long-timescale behaviour of weather, climate and geological systems.