Regensburg 2022 – wissenschaftliches Programm

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

BP: Fachverband Biologische Physik

BP 15: Protein Structure and Single Molecules

BP 15.6: Vortrag

Mittwoch, 7. September 2022, 11:45–12:00, H13

Identifying the Functional Dynamics in Proteins - Divide and Conquer the Feature Space — •Daniel Nagel, Georg Diez, and Gerhard Stock — Biomolecular Dynamics, Institute of Physics, Albert-Ludwigs-Universität, 79104 Freiburg, Germany

The function of proteins is closely linked to their conformational changes. To support experiments, molecular dynamics simulations allow high spatiotemporal resolution while generating large amounts of data. To model and interpret them, it is essential to identify suitable features, such as backbone dihedral angles or interresidual distances. However, in this high-dimensional feature space---in addition to the motion of interest---one finds uncorrelated motions described by small subsets of features, which poses a difficult challenge for the subsequent dimensionality reduction and understanding of the underlying biological process.

In the following we present an effective and scalable correlation-based feature selection method (MoSAIC) that identifies functional dynamics in the feature space and separates it from noise in order to facilitate the further analysis. To demonstrate the different application purposes, we adopt the unsupervised method to systems of various complexity.

G. Diez, D. Nagel, and G. Stock, Correlation-based feature selection to identify functional dynamics in proteins, arxiv:2204.02770, 2022