Regensburg 2022 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

BP: Fachverband Biologische Physik

BP 7: Poster 1

BP 7.7: Poster

Montag, 5. September 2022, 18:00–20:00, P1

Correlation-based feature selection to identify functional dynamics in proteins — •Georg Diez, Daniel Nagel, and Gerhard Stock — Physikalisches Institut, Albert-Ludwigs-Universität Freiburg

Molecular dynamic simulations provide an effective tool for a deeper understanding of proteins and their functioning. In order to shed light on the underlying mechanisms of processes, one typically models the dynamics using some key internal coordinates (or features) which capture the most important conformational changes of the protein. However, one often ends up in a high-dimensional feature space which hampers a straightforward interpretation of the typically very complex dynamics. Adopting the Leiden community detection algorithm [1], we present an effective and scalable approach to divide the feature space into subsets which describe collective motion. By applying this approach to the functional dynamics of different protein systems with varying size, we show that it allows to identify and discard uncorrelated motion and noise. Moreover, it provides an effective dimensionality reduction scheme by extracting the key features, and leads to a detailed understanding of the underlying mechanisms.

[1] Traag et al., "From Louvain to Leiden: guaranteeing well-connected communities", Sci. Rep., 2019