Computational Prediction of Cardiac Electropharmacology - How Much Does the Model Matter?

AbstractAnimal data describing drug interactions in cardiac tissue are abundant, however, nuanced inter-species differences hamper the use of these data to predict drug responses in humans. There are many computational models of cardiomyocyte electrophysiology that facilitate this translation, yet it is unclear whether fundamental differences in their mathematical formalisms significantly impact their predictive power. A common solution to this problem is to perform inter-species translations within a collection of models with internally consistent formalisms, termed a “lineage”, but there has been little effort to translate outputs across lineages. Here, we translate model outputs between lineages from Simula and Washington University for models of ventricular cardiomyocyte electrophysiology of humans, canines, and guinea pigs. For each lineage-species combination, we generated a population of 1000 models by varying common parameters, namely ion conductances, according to a Guassian log-normal distribution with a mean at the parameter’s species-specific default value and standard deviation of 30%.We used partial least squares regression to translate the influences of one model to another using perturbations to calculated descriptors of resulting electrophysiological behavior derived from these parameter variations. Finally, we evaluated translation fidelity by performing a sensitivity analysis between input parameters and output descriptors, as similar sensitivities between models of a common species indicates similar biological mechanisms underlying model behavior. Successful translation between models, especially those from different lineages, will increase confidence in their predictive power.