Dresden 2017 – wissenschaftliches Programm
Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe
BP: Fachverband Biologische Physik
BP 33: Posters - Systems Biology & Gene Expression and Signalling
BP 33.2: Poster
Dienstag, 21. März 2017, 14:00–16:00, P2-OG1
Impact of reparametrization on fitting of ODE models — •Lukas Refisch1, Jens Timmer1,2,3, and Clemens Kreutz1,2 — 1Institute of Physics, University of Freiburg, Freiburg im Breisgau, Germany — 2Center for Biosystems Analysis (ZBSA), University of Freiburg, Freiburg im Breisgau, Germany — 3BIOSS Centre for Biological Signalling Studies, University of Freiburg, Freiburg im Breisgau, Germany
The dynamics of complex biochemical reactions as they occur in living cells can be modeled by ordinary differential equations (ODE). One major task is model calibration, i.e. to estimate parameters like initial concentrations and rate constants based on experimental data. Optimization-based estimation like maximum likelihood is often challenging due to the existence of local minima, the highly nonlinear model responses and the limited precision of numerical ODE solutions.
It has been claimed that parameter transformations are beneficial for fitting of ODE models. Possible parametrizations exploit the model's scaling invariance originating from the free choice of units and the fact, that measurements in molecular cell biology are often taken on a relative scale. However, up to now the impact of reparametrization has not been evaluated in details. For five established models of cellular signaling pathways and infectious diseases with experimental data, we analyze the effect of reparametrization on the performance of optimization. The different influences including the geometry of the likelihood landscape, the choice of initial guesses and the parameter search space are quantified using multivariate statistical analyses.