Abstract
Drug-induced cardiotoxicity remains a predominant driver of compound attrition across non-clinical drug development and post-market safety withdrawals. While the Comprehensive in Vitro Proarrhythmia Assay (CiPA) initiative shifted regulatory screening from an isolated focus on hERG inhibition toward integrated human-relevant platforms, conventional safety pharmacology still contends with the limitations of cardiomyocyte heterogeneity and univariate analytical thresholds. A recent study by Kim et al. (2026) in Archives of Toxicology addresses these challenges by coupling high-purity (>95% cTnT+), ventricular-dominant (78–84% MLC-2V+) human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) with an Artificial Neural Network (ANN), achieving a reported ROC-AUC of 0.982 across the CiPA 28 reference set. Crucially, this integrated bio-computational platform unmasked 'hidden' functional proarrhythmic liabilities in compounds displaying no detectable impedance-based cytotoxicity specifically the tyrosine kinase inhibitors sunitinib and erlotinib which escape detection in standard structural viability assays yet produce severe, time-dependent electrophysiological perturbations. Simultaneously, it provides a continuous, quantitative Torsades de Pointes (TdP) proarrhythmic risk score for refined risk stratification. This Commentary critically evaluates the methodological and translational implications of these findings within the context of the FDA Modernization Act 2.0 and ICH E14/S7B guidelines, dissects the divergence between structural cytotoxicity and functional proarrhythmic liabilities, identifies technical and dataset-size limitations of current machine-learning models, and outlines key standardization requirements needed before this platform can achieve formal regulatory qualification as an accepted New Approach Methodology (NAM).
Keywords
Cardiac safety, hiPSC-derived cardiomyocytes, Artificial intelligence, Anticancer drugs, Proarrhythmia risk, ICH E14/S7B, FDA Modernization Act 2.0, Context of use qualification