You are here:
Computational Approaches for Developing Informative Prior Distributions for Bayesian Calibration of PBPK Models (Book Chapter)
Citation:
DAVIS, J., R. TORNERO-VELEZ, AND R. W. SETZER. Computational Approaches for Developing Informative Prior Distributions for Bayesian Calibration of PBPK Models (Book Chapter). Chapter 18, James B. Knaak, Charles Timchalk, Rogelio Tornero Velez (ed.), Parameters for Pesticide QSAR and PBPK/PD Models for Human Risk Assessment. ACS Publications, Washington, DC, 1099:291-306, (2012).
Impact/Purpose:
Uncertainty in PBPK model parameters and predictions can be assessed via the use of Bayesian statistical methods by combining prior distributions characterizing what is known about the parameters of interest with in vivo data. However, challenges can arise when attempting to quantify parameter (and hence model) uncertainty in PBPK applications with limited or no in vivo data. As stated earlier, these uncertainties can still be assigned in the absence of pharmacokinetic data, but the issue then becomes developing informative prior distributions that truly reflect what is known about a parameter as opposed to either broad weakly informative priors that can overestimate the uncertainty in both the parameters and model predictions or unrealistically precise point priors which erroneously imply certainty. We have demonstrated in this chapter how more informative priors can be developed for chemical-specific PBPK model parameters (e.g., partition coefficients and clearance rates) through the use of appropriate computational predictors (e.g., QSAR and mechanistic models), in vitro methods, and readily available data in the literature. These approaches can be expanded upon and used for other chemical-specific parameters in PBPK models as well as model parameters in other dynamical systems. The priors derived in these approaches can (and are being) used in Bayesian calibration efforts for PBPK models with limited in vivo data and large parameter sets. Ultimately, prior uncertainty in model parameters can be used to more accurately quantify uncertainty in PBPK model predictions.
Description:
Using Bayesian statistical methods to quantify uncertainty and variability in human physiologically-based pharmacokinetic (PBPK) model predictions for use in risk assessments requires prior distributions (priors), which characterize what is known or believed about parameters’ values before observing in vivo data. Experimental in vivo data can then be used in Bayesian calibration of PBPK models to refine priors when it exist. However, when little or no in vivo data are available for calibration efforts, parameter estimates and uncertainties can be obtained from priors. In this chapter, we present approaches for specifying informative priors for chemical-specific PBPK model parameters based on information obtained from chemical structures and in vitro assays. Means and standard deviations (or coefficients of variation) for priors are derived from comparisons of predicted values from computational (e.g., QSAR) methods or in vitro assays and experimentally-determined chemical-specific values for a data set of chemicals.
