DISSECTING CALCIUM RHYTHMS TO PREDICT ARRHYTHMOGENESIS IN HUMAN PLURIPOTENT STEM CELL DERIVED CARDIOMYOCYTES USING A MACHINE LEARNING PLATFORM

Abstract

The assessment of heart electrophysiology to determine arrhythmia susceptibility in vitro remains a challenge. Despite the improvements to various platforms used to study cardiomyocyte electrophysiology, most studies assay for repolarisation blocks as a proxy for proarrhythmic risk. Imperfect arrhythmic risk prediction in patients and drug withdrawals from the clinical market due to undetected cardiotoxicity suggest that the current models are insufficient. To improve on arrhythmogenesis modelling, this work details the utilisation of calcium reporter protein to report on the electrophysiology of cardiomyocytes. We describe a framework to process electrophysiological data into datasets used to train machine learning classifiers to distinguish between healthy and arrhythmic cardiomyocytes accurately. Furthermore, the trained machine learning algorithms can identify the key electrophysiological parameters associated with the studied arrhythmias. We therefore lay the groundwork for high-throughput generation of electrophysiological data suitable for machine learning algorithms to characterise arrhythmogenesis in vitro to complement current preclinical models.