Research
Dynamical reweighting and rare event simulations
Many chemically relevant processes occur on timescales that far exceed the capabilities of conventional molecular dynamics simulations. To address this, numerous methods have been developed to enhance the sampling of these slow molecular events. We contribute to the advancement of two such methods: (1) Girsanov reweighting, which involves biasing the simulation and then re-adjusting the statistical weights of the simulated trajectories to obtain unbiased results, and (2) the square-root approximation of the Fokker-Planck operator, where the continuous exploration of molecular state space is replaced by discretizing the state space and approximating the Fokker-Planck equation on the resulting grid.
Multistate dynamics and Markov models
Chemical systems often exhibit multiple stable conformations, with these states typically separated by a hierarchy of free-energy barriers. Multistate dynamics techniques aim to extract a compact, humanly interpretable model of this complex free-energy landscape from molecular dynamics simulations. Markov State Models (MSMs) are particularly effective for this purpose. They represent the multistate dynamics as a transition matrix, where the eigenvectors and eigenvalues correspond to the system's slow molecular processes, providing insights into the transitions between different states. We contributed to the development of this method and frequently use it to analyze complex molecular dynamics.
More on multistate dynamics and Markov models
Density-based clustering
Density-based clustering is a type of unsupervised machine learning technique used to group data points based on their density in a dataset. This clustering technique is particularly suited to identify conformations in a molecular dynamics data set. We develop the Common-nearest-neightbor algorithm, which can work with irregularly shaped clusters and reliably identifies highly populated regions in data sets with a lot of noise and outliers.
More on density-based clustering
Chemical reactions
Simulating chemical reactions is particularly challenging because they involve bond breaking and formation, which requires either reactive force fields or computationally expensive electronic structure calculations. Additionally, chemical reactions are a prime example of rare events, often necessitating the use of enhanced sampling and reweighting techniques to accurately capture their dynamics.
Allosteric mechanisms in proteins
Allostery is a regulatory mechanism in proteins where binding of a molecule (an allosteric effector) at one site induces a conformational or functional change at a different, often distant, site. This allows the protein to regulate its activity in response to signals. Molecular dynamics (MD) simulations are very powerfull tool for studying allostery, because it gives access to the dynamic behavior and structural changes of proteins over time. By using correlation analyses and mapping the free-energy landscape, we can monitor conformational changes during allosteric transitions and explore the pathways through which these changes propagate from the allosteric site to the active site.
Kinetic models from experiments
More on kinetic models form experiments