BPCPPDYSOE21 – wissenschaftliches Programm

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

CPP: Fachverband Chemische Physik und Polymerphysik

CPP 17: Theorie and Simulation - organized by Jens-Uwe Sommer (Leibniz-Institut für Polymerforschung Dresden, Dresden) (joint session CPP/DY)

CPP 17.1: Hauptvortrag

Mittwoch, 24. März 2021, 09:00–09:40, CPPb

Data-driven methods in polymer physics: exploring the sequence space of copolymers — •Marco Werner — Institut Theorie der Polymere, Leibniz-Institut für Polymerforschung Dresden, Germany

Automated experiments and computer simulation on highly parallel machines push the limits of available data in the field of soft matter. For long polymer chains, however, any data set can cover only a marginal fraction of the giant chemical space and conformation space involved. In this talk, data-driven strategies are discussed that allow to trace hidden physical patterns in both giant spaces by machine learning algorithms. The discussion is centered on the example of hydrophilic / hydrophobic copolymers and their interaction with lipid membranes. A neural network has been trained to predict the free energy landscape near a membrane as a function of the copolymer sequence. The information learned in the hidden neural layers showed that the neural network compressed the sequence space into physically meaningful latent variables. The learned semantics was transferable between simulation data with different levels of coarse-graining, and allowed for a physics-informed inverse search for the copolymer sequence leading to the smallest translocation time through the membrane.