Parameter subset selection techniques for problems in mathematical biology

Johnny T. Ottesen, Christian Haargaard-Olsen, Ralf Smith, Mette Olufsen

Publikation: Bidrag til tidsskriftTidsskriftartikelForskning

Resumé

Patient-specific models for diagnostics and treatment planning require reliable parameter estimation and model predictions. Mathematical models of physiological systems are often formulated as systems of nonlinear ordinary differential equations with many parameters and few options for measuring all state variables. Consequently, it can be difficult to determine which parameters can reliably be estimated from available data. This investigation highlights pitfalls associated with practical parameter identifiability and subset selection. The latter refer to the process associated with selecting a subset of parameters that can be identified uniquely by parameter estimation protocols. The methods will be demonstrated using five examples of increasing complexity, as well as with patient-specific model predicting arterial blood pressure. This study demonstrates that methods based on local sensitivities are preferable in terms of computational cost and model fit when good initial parameter values are available, but that global methods should be considered when initial parameter value is not known or poorly understood. For global sensitivity analysis, Morris screening provides results in terms of parameter sensitivity ranking at a much lower computational cost.
OriginalsprogEngelsk
TidsskriftBiological Cybernetics
Sider (fra-til)1-18
Antal sider19
ISSN0340-1200
DOI
StatusE-pub ahead of print - 30 okt. 2018

Emneord

    Citer dette

    Ottesen, Johnny T. ; Haargaard-Olsen, Christian ; Smith, Ralf ; Olufsen, Mette. / Parameter subset selection techniques for problems in mathematical biology. I: Biological Cybernetics. 2018 ; s. 1-18.
    @article{ca18908fe96442138ac67c802169a118,
    title = "Parameter subset selection techniques for problems in mathematical biology",
    abstract = "Patient-specific models for diagnostics and treatment planning require reliable parameter estimation and model predictions. Mathematical models of physiological systems are often formulated as systems of nonlinear ordinary differential equations with many parameters and few options for measuring all state variables. Consequently, it can be difficult to determine which parameters can reliably be estimated from available data. This investigation highlights pitfalls associated with practical parameter identifiability and subset selection. The latter refer to the process associated with selecting a subset of parameters that can be identified uniquely by parameter estimation protocols. The methods will be demonstrated using five examples of increasing complexity, as well as with patient-specific model predicting arterial blood pressure. This study demonstrates that methods based on local sensitivities are preferable in terms of computational cost and model fit when good initial parameter values are available, but that global methods should be considered when initial parameter value is not known or poorly understood. For global sensitivity analysis, Morris screening provides results in terms of parameter sensitivity ranking at a much lower computational cost.",
    keywords = "Modeling, Parameter estimation, Parameter identifiability, Parameter subset selection",
    author = "Ottesen, {Johnny T.} and Christian Haargaard-Olsen and Ralf Smith and Mette Olufsen",
    year = "2018",
    month = "10",
    day = "30",
    doi = "10.1007/s00422-018-0784-8",
    language = "English",
    pages = "1--18",
    journal = "Biological Cybernetics",
    issn = "0340-1200",
    publisher = "Physica-Verlag",

    }

    Parameter subset selection techniques for problems in mathematical biology. / Ottesen, Johnny T.; Haargaard-Olsen, Christian; Smith, Ralf; Olufsen, Mette.

    I: Biological Cybernetics, 30.10.2018, s. 1-18.

    Publikation: Bidrag til tidsskriftTidsskriftartikelForskning

    TY - JOUR

    T1 - Parameter subset selection techniques for problems in mathematical biology

    AU - Ottesen, Johnny T.

    AU - Haargaard-Olsen, Christian

    AU - Smith, Ralf

    AU - Olufsen, Mette

    PY - 2018/10/30

    Y1 - 2018/10/30

    N2 - Patient-specific models for diagnostics and treatment planning require reliable parameter estimation and model predictions. Mathematical models of physiological systems are often formulated as systems of nonlinear ordinary differential equations with many parameters and few options for measuring all state variables. Consequently, it can be difficult to determine which parameters can reliably be estimated from available data. This investigation highlights pitfalls associated with practical parameter identifiability and subset selection. The latter refer to the process associated with selecting a subset of parameters that can be identified uniquely by parameter estimation protocols. The methods will be demonstrated using five examples of increasing complexity, as well as with patient-specific model predicting arterial blood pressure. This study demonstrates that methods based on local sensitivities are preferable in terms of computational cost and model fit when good initial parameter values are available, but that global methods should be considered when initial parameter value is not known or poorly understood. For global sensitivity analysis, Morris screening provides results in terms of parameter sensitivity ranking at a much lower computational cost.

    AB - Patient-specific models for diagnostics and treatment planning require reliable parameter estimation and model predictions. Mathematical models of physiological systems are often formulated as systems of nonlinear ordinary differential equations with many parameters and few options for measuring all state variables. Consequently, it can be difficult to determine which parameters can reliably be estimated from available data. This investigation highlights pitfalls associated with practical parameter identifiability and subset selection. The latter refer to the process associated with selecting a subset of parameters that can be identified uniquely by parameter estimation protocols. The methods will be demonstrated using five examples of increasing complexity, as well as with patient-specific model predicting arterial blood pressure. This study demonstrates that methods based on local sensitivities are preferable in terms of computational cost and model fit when good initial parameter values are available, but that global methods should be considered when initial parameter value is not known or poorly understood. For global sensitivity analysis, Morris screening provides results in terms of parameter sensitivity ranking at a much lower computational cost.

    KW - Modeling

    KW - Parameter estimation

    KW - Parameter identifiability

    KW - Parameter subset selection

    UR - https://link.springer.com/content/pdf/10.1007%2Fs00422-018-0784-8.pdf

    U2 - 10.1007/s00422-018-0784-8

    DO - 10.1007/s00422-018-0784-8

    M3 - Journal article

    SP - 1

    EP - 18

    JO - Biological Cybernetics

    JF - Biological Cybernetics

    SN - 0340-1200

    ER -