Strategies for parameter uncertainty analysis

J. Vanlier, C.A. Tiemann, J.A.L. Jeneson, P.A.J. Hilbers, N.A.W. Riel, van

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

Abstract

Models of biochemical pathways often comprise of hundreds of parameters. Since in many cases the available data is limited many parameter values are unknown. Moreover, this data is hampered by noise, scaling and offset parameters which reduce the inferential power of the data. As a consequence, the modeller is left with a parameter estimation problem, where multiple parameter sets can adequately describe the acquired data to an acceptable degree [1,2].In recent years, several new methods to deal with inferential problems have been developed, both from the Bayesian as well as the frequentist side. Our work identifies opportunities to combine different approaches into a consistent strategy for uncertainty analysis. Methods The inferential methods we employ include Multiple Minimisation approaches, Profile Likelihood analysis [2] and Hessian-based Markov Chain Monte Carlo [1]. These methods were subsequently used to parameterise a 7 parameter mass action model of the JAK-STAT pathway [2].Results Results indicate that combining several approaches leads to insights into model behaviour and avoids potential pitfalls of each individual method. Depending on the data used for parameterisation some parameters can be quite constrained by data, while others are not. In such cases, assuming that the model can be described by a single parameter set can lead to overconfident conclusions. Exploring the effects of uncertainty on model predictions to be able to assess whether the predictions of interest can truly be used to falsify a given hypothesis is therefore important. Furthermore, it appeared that model parameters and parameters involved in scaling the data were strongly related to kinetic model parameters and can hamper model analysis. Our results provide suggestions for dealing with such situations.
LanguageEnglish
Title of host publicationPresentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany
Place of PublicationHeidelberg
StatePublished - 2011

Fingerprint

uncertainty analysis
parameter
Markov chain
prediction
parameterization
method
kinetics

Cite this

Vanlier, J., Tiemann, C. A., Jeneson, J. A. L., Hilbers, P. A. J., & Riel, van, N. A. W. (2011). Strategies for parameter uncertainty analysis. In Presentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany Heidelberg.
Vanlier, J. ; Tiemann, C.A. ; Jeneson, J.A.L. ; Hilbers, P.A.J. ; Riel, van, N.A.W./ Strategies for parameter uncertainty analysis. Presentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany. Heidelberg, 2011.
@inproceedings{6ad5efefdff94165a903442f74ea3e22,
title = "Strategies for parameter uncertainty analysis",
abstract = "Models of biochemical pathways often comprise of hundreds of parameters. Since in many cases the available data is limited many parameter values are unknown. Moreover, this data is hampered by noise, scaling and offset parameters which reduce the inferential power of the data. As a consequence, the modeller is left with a parameter estimation problem, where multiple parameter sets can adequately describe the acquired data to an acceptable degree [1,2].In recent years, several new methods to deal with inferential problems have been developed, both from the Bayesian as well as the frequentist side. Our work identifies opportunities to combine different approaches into a consistent strategy for uncertainty analysis. Methods The inferential methods we employ include Multiple Minimisation approaches, Profile Likelihood analysis [2] and Hessian-based Markov Chain Monte Carlo [1]. These methods were subsequently used to parameterise a 7 parameter mass action model of the JAK-STAT pathway [2].Results Results indicate that combining several approaches leads to insights into model behaviour and avoids potential pitfalls of each individual method. Depending on the data used for parameterisation some parameters can be quite constrained by data, while others are not. In such cases, assuming that the model can be described by a single parameter set can lead to overconfident conclusions. Exploring the effects of uncertainty on model predictions to be able to assess whether the predictions of interest can truly be used to falsify a given hypothesis is therefore important. Furthermore, it appeared that model parameters and parameters involved in scaling the data were strongly related to kinetic model parameters and can hamper model analysis. Our results provide suggestions for dealing with such situations.",
author = "J. Vanlier and C.A. Tiemann and J.A.L. Jeneson and P.A.J. Hilbers and {Riel, van}, N.A.W.",
year = "2011",
language = "English",
booktitle = "Presentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany",

}

Vanlier, J, Tiemann, CA, Jeneson, JAL, Hilbers, PAJ & Riel, van, NAW 2011, Strategies for parameter uncertainty analysis. in Presentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany. Heidelberg.

Strategies for parameter uncertainty analysis. / Vanlier, J.; Tiemann, C.A.; Jeneson, J.A.L.; Hilbers, P.A.J.; Riel, van, N.A.W.

Presentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany. Heidelberg, 2011.

Research output: Chapter in Book/Report/Conference proceedingConference contributionAcademicpeer-review

TY - GEN

T1 - Strategies for parameter uncertainty analysis

AU - Vanlier,J.

AU - Tiemann,C.A.

AU - Jeneson,J.A.L.

AU - Hilbers,P.A.J.

AU - Riel, van,N.A.W.

PY - 2011

Y1 - 2011

N2 - Models of biochemical pathways often comprise of hundreds of parameters. Since in many cases the available data is limited many parameter values are unknown. Moreover, this data is hampered by noise, scaling and offset parameters which reduce the inferential power of the data. As a consequence, the modeller is left with a parameter estimation problem, where multiple parameter sets can adequately describe the acquired data to an acceptable degree [1,2].In recent years, several new methods to deal with inferential problems have been developed, both from the Bayesian as well as the frequentist side. Our work identifies opportunities to combine different approaches into a consistent strategy for uncertainty analysis. Methods The inferential methods we employ include Multiple Minimisation approaches, Profile Likelihood analysis [2] and Hessian-based Markov Chain Monte Carlo [1]. These methods were subsequently used to parameterise a 7 parameter mass action model of the JAK-STAT pathway [2].Results Results indicate that combining several approaches leads to insights into model behaviour and avoids potential pitfalls of each individual method. Depending on the data used for parameterisation some parameters can be quite constrained by data, while others are not. In such cases, assuming that the model can be described by a single parameter set can lead to overconfident conclusions. Exploring the effects of uncertainty on model predictions to be able to assess whether the predictions of interest can truly be used to falsify a given hypothesis is therefore important. Furthermore, it appeared that model parameters and parameters involved in scaling the data were strongly related to kinetic model parameters and can hamper model analysis. Our results provide suggestions for dealing with such situations.

AB - Models of biochemical pathways often comprise of hundreds of parameters. Since in many cases the available data is limited many parameter values are unknown. Moreover, this data is hampered by noise, scaling and offset parameters which reduce the inferential power of the data. As a consequence, the modeller is left with a parameter estimation problem, where multiple parameter sets can adequately describe the acquired data to an acceptable degree [1,2].In recent years, several new methods to deal with inferential problems have been developed, both from the Bayesian as well as the frequentist side. Our work identifies opportunities to combine different approaches into a consistent strategy for uncertainty analysis. Methods The inferential methods we employ include Multiple Minimisation approaches, Profile Likelihood analysis [2] and Hessian-based Markov Chain Monte Carlo [1]. These methods were subsequently used to parameterise a 7 parameter mass action model of the JAK-STAT pathway [2].Results Results indicate that combining several approaches leads to insights into model behaviour and avoids potential pitfalls of each individual method. Depending on the data used for parameterisation some parameters can be quite constrained by data, while others are not. In such cases, assuming that the model can be described by a single parameter set can lead to overconfident conclusions. Exploring the effects of uncertainty on model predictions to be able to assess whether the predictions of interest can truly be used to falsify a given hypothesis is therefore important. Furthermore, it appeared that model parameters and parameters involved in scaling the data were strongly related to kinetic model parameters and can hamper model analysis. Our results provide suggestions for dealing with such situations.

M3 - Conference contribution

BT - Presentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany

CY - Heidelberg

ER -

Vanlier J, Tiemann CA, Jeneson JAL, Hilbers PAJ, Riel, van NAW. Strategies for parameter uncertainty analysis. In Presentation at the 5th Workshop on Monte Carlo Method, 14 January 2011, Heidelberg, Germany. Heidelberg. 2011.