Identifiability of parameters in PBPK models
Ladda ner
Typ
Examensarbete för masterexamen
Master Thesis
Master Thesis
Program
Engineering mathematics and computational science (MPENM), MSc
Publicerad
2019
Författare
Watanabe, Simon Berglund
Modellbyggare
Tidskriftstitel
ISSN
Volymtitel
Utgivare
Sammanfattning
Inthefieldofpharmacologics,physiologically-basedpharmacokinetic(PBPK)models can be used for predicting the pharmacokinetics of a drug compound in the body. These models are often a system of ordinary differential equations (ODEs) that describe the transport of a drug between different compartments of the body. The models depend on several parameters, some of which cannot be measured experimentally and instead these parameters are often estimated from experimental data using maximum likelihood. However, in many applications in systems biology, estimates will suffer from unidentifiability issues, meaning that well-determined estimates cannot be inferred from the data [17]. Thisproblemcomesintwoforms,structuralunidentifiabilityandpracticalunidentifiability,bothofwhichcanbeanalyzedwiththeprofilelikelihoodmethoddeveloped by Raue et.al [14]. The profile likelihood method is a numerical method for calculating likelihood-based confidence intervals of the parameters, which are then used to assess identifiability. In this project the profile likelihood method is implemented in MATLAB and used to perform identifiability analysis on key model parameters for three PBPK models using simulated data. Thus, the results of this project are both a showcase of the profilelikelihoodmethodandananalysisoftheidentifiabilityofparametersinsome specific models used for pulmonary drug delivery. The results indicate that if very precise measurements could be taken then all parameters considered would be identifiable. When a reasonable measurement error is applied on the simulated data the same is not true. Some parameters, such as the in-vivo pulmonary permeability and deposition fraction will remain identifiable, but most other parameters will suffer from practical unidentifiability. With a reasonable measurement error the identifiability of most model parameters will also be dependent on the particular error realization. To address these issues, additional dataisconsideredbyobservinghowtheuncertaintyinparameterestimatesimpacts observables. Bythismethod(alsosuggestedbyRaue[14])additionalmeasurements are introduced in an effective manner to potentially resolve unidentifiabilities. Keywords: structural unidentifiability, practical unidentifiability, pulmonary drug delivery, maximum likelihood estimation.
Beskrivning
Ämne/nyckelord
Matematik , Mathematics