Evaluating the predictive accuracy of ion-channel models using data from multiple experimental designs

Mathematical models are increasingly being relied upon to provide quantitatively accurate predictions of cardiac electrophysiology. Many such models concern the behaviour of particular subcellular components (namely, ion channels) which, together, allow the propagation of electrical signals through heart-muscle tissue; that is, the firing of action potentials. In particular, I_Kr, a voltage-sensitive potassium ion-channel current, is of interest owing to the central pore of its primary protein having a propensity to blockage by various small molecules. We use newly collected data obtained from an ensemble of voltage-clamp experiment designs (protocols) to validate the predictive accuracy of various dynamical models of I_Kr. To do this, we fit models to each protocol individually and quantify the error in the resultant model predictions for other protocols. This allows the comparison of predictive accuracy for I_Kr models under a diverse collection of previously unexplored dynamics. Our results highlight heterogeneity between parameter estimates obtained from different cells, suggesting the presence of latent effects not yet accounted for in our models. This heterogeneity has a significant effect on our parameter estimates and suggests routes for model improvement.This article is part of the theme issue 'Uncertainty quantification for healthcare and biological systems (Part 1)'.