In Silico Drug Action Estimation from Cardiac Action Potentials by Model Fitting in a Sampled Parameter Space

Abstract

Model-based predictive approaches have been receiving increasing attention as a valuable tool to reduce high development cost in drug discovery. Recently developed cardiac cell models are integrated with major ion channels and capable of reproducing action potentials (AP) precisely. In this work, a model-fitting-based approach for estimating drug action from cardiac AP recordings is investigated. Giving a test AP, the activities of involved ion channels can be determined by fitting the cell model to reproduce the test AP. Using experimental APs recordings both before and after drug dose, drug actions can be estimated by changes in activity of corresponding ion channels. Local-gradient-based optimization methods are too time-consuming due to the high computational cost of cardiac cell models. A fast approach using only pre-calculated samples to improve computational efficiency is addressed. The searching strategy in the sampled space is divided into two steps: in the first step, the sample of best similarity comparing with the test AP is selected; then response surface approximation using the neighboring samples is followed and the estimation value is obtained by the approximated surface. This approach showed quite good estimation accuracy for a large number of simulation tests. Also results for animal AP recordings from drug dose trials were exemplified in which case the ICaL inhibition effect of Nifedipine and IKr inhibition effect of E4031 were correctly discovered.