Enhancement of Skin Permeability Prediction through PBPK Modeling, Bayesian Inference, and Experiment Design

Abstract

Physiologically based pharmacokinetic (PBPK) models of skin absorption are a powerful resource for estimating drug delivery and chemical risk of dermatological products. This paper presents a PBPK workflow for the quantification of the mechanistic determinants of skin permeability and the use of these quantities in the prediction of skin absorption in novel contexts. A state-of-the-art mechanistic model of dermal absorption was programmed into an open-source modeling framework. A sensitivity analysis was performed to identify the uncertain compound-specific, individual-specific, and site-specific model parameters that impact permeability. A Bayesian Markov Chain Monte Carlo algorithm was employed to derive distributions of these parameters given in vitro experimental permeability measurements. Extrapolations to novel contexts were generated by simulating the model following its update with samples drawn from the learned distributions as well as parameters that represent the intended scenario. This algorithm was applied multiple times, each using a unique set of permeability measurements sourced under experimental contexts that differ in terms of the compound, vehicle pH, skin sample anatomical site, and the number of compounds under which each subject’s skin samples were tested. Among the data sets used in this study, the highest accuracy and precision in the extrapolated permeability was achieved in those that include measurements conducted under multiple vehicle pH levels and in which individual subjects’ skin samples are tested under multiple compounds. This work thus identifies factors for consideration in the design of experiments for the purpose of training dermal models to robustly estimate drug delivery and chemical risk.