Integrating high-fidelity hiPSC-cardiomyocytes with AI-driven modeling for enhanced proarrhythmic risk assessment

Abstract

Cardiotoxicity remains the leading driver of drug attrition; however, its prediction remains suboptimal when conventional hERG assays and animal models are used. Human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) offer a human-relevant alternative that aligns with the CiPA initiative and ICH E14/S7B guidelines. This study validated an integrated platform that combines high-purity hiPSC-CMs with Artificial Intelligence (AI) to enhance the accuracy of predicting drug-induced proarrhythmic risk. Phenotypic characterization of the hiPSC-CMs demonstrated high cardiac differentiation efficiency (cTnT + > 95%) and a predominant ventricular-like identity (MLC-2 V + , 78–84%), ensuring biological relevance for ventricular arrhythmia assessment. Electrophysiological data from 28 CiPA reference compounds were collected via Multielectrode Array (MEA) to train multiple machine learning models. The Artificial Neural Network outperformed the other architectures, achieving a superior ROC-AUC of 0.982. The utility of the platform was evaluated using 12 anticancer agents. Although most drugs showed dose-dependent reductions in impedance-based viability, four compounds (Idarubicin, Erlotinib, Sunitinib, Cyclophosphamide) did not exhibit overt structural cytotoxicity. However, MEA analysis revealed significant functional perturbations, including FPDcF prolongation, in sunitinib- and erlotinib-treated samples after long-term treatment. The AI model subsequently classified these two agents as high-to-intermediate risk for Torsades de Pointes (TdP), thereby quantifying their time-dependent proarrhythmic liabilities. These findings show the platform’s ability to detect hidden functional cardiotoxicity, often missed by standard viability assays. The AI-hiPSC-CM system offers a high-throughput, early-stage safety screening tool that bridges in vitro data and clinical outcomes with a standardized risk assessment framework.

Bibliographic Information