Dresden 2020 – wissenschaftliches Programm
Die DPG-Frühjahrstagung in Dresden musste abgesagt werden! Lesen Sie mehr ...
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
BP: Fachverband Biologische Physik
BP 19: Poster VII
BP 19.11: Poster
Dienstag, 17. März 2020, 14:00–16:00, P2/3OG
Metaheuristic Optimization of Biomolecular Simulation Parameters — •Marie Weiel1, Markus Götz1, and Alexander Schug2 — 1Karlsruhe Institute of Technology, Karlsruhe, Germany — 2Jülich Supercomputing Centre, Jülich, Germany
Owing to the structure-function paradigm, a wealth of structural information on proteins has been accumulated. Experimental data might be ambiguous and have to be interpreted to access their actual content. A common way is to integrate them into molecular simulations via an energetic bias favoring conformations concordant with the data. Data-assisted simulations often rely on parameters, the choice of which is far from trivial but crucial for performance. A central question is how to weight experimental information with respect to prior knowledge in the underlying physical model. We propose a metaheuristic particle-swarm based optimization method. In our setup, a particle corresponds to a simulation using a particular combination of parameters to be optimized. To assess simulation outcome, the ensemble's physicality and its agreement with the data are considered. We use Rosetta and bias energy, respectively, out of which the objective function is constructed. To assign equal importance, each contribution has to be rescaled to the same order of magnitude. As the related objective function hyperparameters are a priori unknown but have a critical impact on the search-space topology, they are refined on the fly while the algorithm proceeds. The better different contribution ranges are known from complete simulations, the more accurate values can be chosen, improving optimization efficiency and thus simulation quality.