CETO: A Novel in silico Cardiac Safety Assessment Using Action Potential Shape-Derived Torsadogenic Safety Margins
Year: 2026
Type: dataset
Abstract: This study presents an in silico framework that combines deep learning-driven action potential analysis with exposure-informed pharmacodynamic modeling to predict drug-induced Torsades de Pointes (TdP) risk. Using the CiPAORdv1.0 model, action potentials were simulated for 28 CiPA reference compounds and subsequently classified by a convolutional neural network (CNN) as either arrhythmic or non-arrhythmic. The resulting probability profiles across drug concentrations were then fitted with sigmoid dose-response curves to estimate EC50 values and calculate the Torsadogenic Safety Margin (TSM) based on clinical Cmax. These TSM values were subsequently incorporated into an ordinal logistic regression (OLR) model to categorize compounds into high-, intermediate-, or low-risk TdP groups. Model performance was evaluated through 10,000 randomized testing iterations under two distinct assessment settings. The CNN demonstrated strong accuracy in distinguishing arrhythmic from non-arrhythmic waveforms, while the TSM-based OLR showed excellent ordinal classification performance, with both AUC1 and AUC2 exceeding 0.94 and likelihood ratios supporting clear separation among risk categories, thereby matching or outperforming previously published CiPA-based approaches. Overall, this hybrid in silico strategy unifies electrophysiological waveform characteristics, dose-response relationships, and clinical exposure into a single decision-making framework, offering an interpretable and reliable tool for large-scale, model-informed cardiac safety evaluation during drug development. The uploaded file contains the supplementary materials, including the detailed mathematical formulation of the CETO framework.
Language: English
Subfield: Cardiology and Cardiovascular Medicine
Field: Medicine
Domain: Health Sciences
Open Access status: green
OpenAlex ID: W7152604669