SKM 2023 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
BP: Fachverband Biologische Physik
BP 3: Computational Biophysics I
BP 3.6: Vortrag
Montag, 27. März 2023, 11:15–11:30, BAR 0106
Clustering Molecular Dynamics Trajectories using Density and Flux — •Jayashrita Debnath1 and Gerhard Hummer1,2 — 1Max Planck Institute of Biophysics, Frankfurt am Main, Germany — 2Goethe University, Frankfurt am Main, Germany
Molecular dynamics (MD) simulations are a powerful tool for studying a wide range of molecular systems and processes, with applications ranging from materials science to biology and medicine. Analyzing these simulations often involves finding a low-dimensional representation of the trajectory data and clustering the sampled configurations into kinetically relevant metastable states. The steady growth in the time and length scales of MD simulations, and in the complexity of their molecular systems, necessitates the development of new analysis tools that do not rely entirely on chemical or physical intuition. Here, we propose a neural network based unsupervised algorithm that can identify states using the static and dynamic information encoded in the trajectories. The network identifies metastable states by modeling a probability distribution of the data in a reduced dimensional space and learns the state boundaries by minimizing the flux between states. Furthermore, it can learn the optimal number of states from single long equilibrium trajectories or multiple short ones. After demonstrating the effectiveness of this method for a toy potential, we apply it to trypsin-benzamidine unbinding as a model of drug binding kinetics, to and folding-unfolding transitions of the villin headpiece subdomain.