Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
BP: Fachverband Biologische Physik
BP 3: Computational Biophysics I
BP 3.8: Vortrag
Montag, 27. März 2023, 11:45–12:00, BAR 0106
Finding pathways in molecular dynamics simulations using machine learning and graph methods — •Steffen Wolf1, Miriam Jäger1, Victor Tänzel1, Simon Bray1,2, Matthias Post1, and Gerhard Stock1 — 1Biomolecular Dynamics, Institute of Physics, University of Freiburg, 79104 Freiburg, Germany — 2Bioinformatics Group, Institute of Informatics, University of Freiburg, 79110 Freiburg, Germany
Understanding the mechanisms of biomolecular systems and complexes, e.g., of protein-ligand (un)binding, requires the understanding of paths such systems take between metastable states. In MD simulation data, paths are usually not observable per se, but need to be inferred from simulation trajectories. Here we present novel approaches to cluster trajectories according to similarities. These approaches include neighbor-nets allowing to correct for input data ambiguity [1] and an unsupervised learning approach employing only a single free parameter [2]. We demonstrate how such clusters of trajectories correspond to pathways, and how the approaches help in the identification of reaction coordinates for a considered process. Last, we present a theoretical framework how potentials of mean force can be calculated for individual pathways, and how these potentials and kinetics along paths can be combined into a comprehensive complete free energy profile and process kinetics.
[1] Bray, S., Tänzel, V. & Wolf, S. J. Chem. Inf. Model. 62, 4591-4604 (2022). [2] Diez, G., Nagel, D. & Stock, G. J. Chem. Theory Comput. 18, 5079-5088 (2022).