Coupled Problems in Biomechanics and Systems Biology
Monday, 26 March 2018
2:00pm - 3:30pm
Campus Stuttgart-Vaihingen, Pfaffenwaldring 47
Room 47.03
On the Parameterisation of Multi-scale Models in Life Sciences
Jan Hasenauer, Helmholtz Center Munich
Mechanistic computational models are powerful tools in modern life sciences. Similar to experimental techniques, mechanistic models facilitate an assessment of hypothesis testing. To achieve this, the unknown model parameters have to be estimated from experimental data. For stochastic processes on multiple time and length scales, parameter estimation is often still intractable. Here, we present established Approximate Bayesian Computation (ABC) - Sequential Monte Carlo (SMC) methods which are conceptually applicable for the calibration of stochastic multi-scale models but computationally demanding. To facilitate the application of these methods to computationally intensive models, we introduce pyABC, an open-source Python toolbox with high-performance computing (HPC) capabilities which offers several features enhancing computational efficiency. We demonstrate pyABC using multi-scale models for tumor growth and immune response. For both applications, pyABC achieves better results than manual tuning and enables an assessment of parameter and prediction uncertainties.
Inference of Finite Micture Models with Binned Single Cell Data
Nicole Radde, University of Stuttgart
Finite mixture models have successfully been used to analyze a variety of biological data. Recent examples include single cell data and the quantification of cell-to-cell heterogeneity or the classification of cancer subtypes. Calibration of these models to experimental data is challenging for uncensored data and also for binned data, as often available in practice.
Here we investigate the problem of parameter estimation and model selection for finite mixture models from a theoretical perspective and on a real case study with time lapse microscopy data. In statistics it has long been known that the calibration of mixture models constitutes an ill-posed problem. We illustrate this fact on mixtures of different distribution families and discuss the effect of binning on this inverse problem. We demonstrate that a proper treatment of binning can in fact facilitate estimation of the number of mixtures compared to inference from uncensored data, an at first glance surprising result.
Methods and Models to Study Biomechanics and Motor Control
Syn Schmitt, University of Stuttgart
The dynamics of a biological system is truly unique. No other system, even highly sophisticated, engineered structures do not show the same inherent complexity neither in their structure nor in their functioning. To understand biological systems better, models could come into play. In this talk, methods and models are presented to study biological locomotion. Spanning multiple spatial and temporal scales, being developed over a long time of evolution, and having data-poor and datarich regimes available for validation, modelling a biological system proves a demanding challenge. However, being able to simulate a biological motion using neuro-musculo-skeletal models opens up the space to answer questions like, as to whether non-linearities in biology are a work around due to available building materials or a design feature to enable simpler motion planning and execution?
Uncertainty Quantification of Multi-X Liver Lobule Damage Simulation
Tim Ricken, University of Stuttgart
The human liver regulates metabolism in a complex time depending and non-linear coupled functionperfusion-mechanism. The viability of the organ is affected by a failure in the liver structure, e.g. lipid accumulation. We present a computational multi-X model for the human liver which is composed of three coupled submodels for the organ, lobule- and cell-scale. Thus, the effect of inhomogenously distributed growing fat vacuoles in the liver cells could be computed. To address uncertainty quantification, two promising approaches of analytical and stochastic sensitivity analysis will enhance the deterministic structural analysis. The variational sensitivity analysis is used to capture the impact of different parameters as continuous functions. An advantage is the accurate approximation of the solution space and the efficient computation time. In the probabilistic sensitivity analysis, we use Bayes statistics will enable to receive accurate information with just a few simulations.
Postertrack contributions:
Karsten Kuritz (PP4.063); Ewa Anna Oprzeska-Zingrebe (PP4.064); Jonas Landsgesell & Christian Holm & Patrick Kreissl & Georg Rempfer & Florian Weik & Kai Szuttor (PP4.065); Wolfgang Halter (PP4.066); Stefan Engblom (PP4.067); Simon Wolfen (PP.068); Ekin Altan & Leonardo Gizzi & Sergey Oladyshkin & Oliver Röhrle (PP4.069); Tobias Koeppl & Paolo Zunino & Barbara Wohlmuth & Ettore Vidotto (PP4.071); Daniela Stöhr (PP4.072); Dirke Imig (PP4.073); Frank Maier (PP4.074); Davina Fink (PP4.075); Sergio Morales & Oliver Röhrle (PP4.076); Patrick Schröder & Arndt Wagner & Daniela Stöhr & Rehm Markus & Wolfgang Ehlers (PP4.077); Debdas Paul & Nicole Radde; Sergio Morales & Oliver Röhrle (PP4.079); Maria Hammer & Michael Günther & Daniel F. B. Haeufle & Johannes Walter & Syn Schmitt (PP4.080); Antje Jensch & Katharina Bitschar & Nicole Radde & Monilola Olayioye (PP4.081); Nehzat Emamy & Thomas Ertl & Dominik Göddeke & Thomas Klotz & Aaron Krämer & Michael Krone & Benjamin Maier & Miriam Mehl & Tobias Rau & Oliver Röhrle (PP4.082); Christian Bleiler (PP4.083); Klaudius Scheufele & Andreas Mang (PP4.084); Katrin Stollenmaier (PP4.085).
