Interpretable machine learning of action potential duration restitution properties in single-cell models of atrial cardiomyocyte

Abstract

Action potential duration (APD) restitution curve and its maximal slope (Smax) reflect single cell-level instability for inducing chaotic heart rhythms. Conventional parameter sensitivity analysis often fails to describe nonlinear relationships between ion channel parameters and electrophysiological phenotypes, such as Smax. We explored the parameter-phenotype mapping in in silico atrial cell models through interpretable machine learning (ML) approaches. We generated a population of 5,000 single-cell atrial cell models by log-uniformly sampling parameter combinations with the Sobol sequence scheme.