TY - CONF
T1 - Comparative study of global parameter Sensitivity Analyses of models based on Ordinary Differential Equations
AU - Yang, H.
AU - Vanlier, J.
AU - Hilbers, P.A.J.
AU - Riel, van, N.A.W.
N1 - Poster presented at the 3rd Dutch Bio-Medical Engineering Conference (BME 2011), 20-21 January 2011, Egmond aan Zee, The Netherlands
PY - 2011
Y1 - 2011
N2 - Biomedical engineering research often involves simulating the dynamic behavior of mathematical models composed of Ordinary Differential Equations (ODEs). However, to be able to simulate the model all parameters, including initial conditions, must be experimentally measured or inferred to specify the model. In silico Parameter Sensitivity Analysis (PSA) serves as a cornerstone in the analysis of complex systems. It provides modelers a means to identify the most critical parameters with respect to the model output and, thereby, to focus the design of experiments. When precise values of parameters are unknown, Global Parameter Sensitivity Analysis (GPSA), which explores a large part of the parameter space, including the interaction of uncertainty in many parameters, is preferred to Local Parameter Sensitivity Analysis (LPSA). Though various GPSA methods have been proposed in last decades [1], there is still a demand for a comparative study of different GPSA methods in order to elucidate the information they provide as well as their inherent limitations. In this work, we focus on the comparison of several GPSA methods for systems based on ODEs, and our objective is to investigate the applicability of such methods and to assess whether the results are reliable. Three different GPSA methods are applied to time varying outputs in multiple ODEs models based on mass action kinetics, some of which are nonlinear, non-monotonic and non-additive; furthermore the influence of the assumed uncertainty distribution is investigated. Results are compared to reveal more model insights than that from individual GPSA analyses. It is shown that in a model based on ODEs, the GPSA results are influenced by four factors: the way sensitivity is defined, ODEs model structure, information extraction function, and uncertainty distributions of parameters. Additionally, we propose a general strategy for employing sensitivity analysis in order to obtain a more reliable and informative understanding of the models.
AB - Biomedical engineering research often involves simulating the dynamic behavior of mathematical models composed of Ordinary Differential Equations (ODEs). However, to be able to simulate the model all parameters, including initial conditions, must be experimentally measured or inferred to specify the model. In silico Parameter Sensitivity Analysis (PSA) serves as a cornerstone in the analysis of complex systems. It provides modelers a means to identify the most critical parameters with respect to the model output and, thereby, to focus the design of experiments. When precise values of parameters are unknown, Global Parameter Sensitivity Analysis (GPSA), which explores a large part of the parameter space, including the interaction of uncertainty in many parameters, is preferred to Local Parameter Sensitivity Analysis (LPSA). Though various GPSA methods have been proposed in last decades [1], there is still a demand for a comparative study of different GPSA methods in order to elucidate the information they provide as well as their inherent limitations. In this work, we focus on the comparison of several GPSA methods for systems based on ODEs, and our objective is to investigate the applicability of such methods and to assess whether the results are reliable. Three different GPSA methods are applied to time varying outputs in multiple ODEs models based on mass action kinetics, some of which are nonlinear, non-monotonic and non-additive; furthermore the influence of the assumed uncertainty distribution is investigated. Results are compared to reveal more model insights than that from individual GPSA analyses. It is shown that in a model based on ODEs, the GPSA results are influenced by four factors: the way sensitivity is defined, ODEs model structure, information extraction function, and uncertainty distributions of parameters. Additionally, we propose a general strategy for employing sensitivity analysis in order to obtain a more reliable and informative understanding of the models.
M3 - Poster
ER -