Bridging mechanistic models and AI for next-generation cardiac safety trials: a loperamide overexposure case study

Drug-induced QT interval prolongation remains a leading indicator of proarrhythmic risk and a major challenge in cardiac safety pharmacology. While regulatory guidelines (ICH S7B/E14) call for improved non-clinical methods [2], mechanistic in silico models offer a powerful yet underused tool for early safety evaluation.

This work aims to present an AI-enhanced framework that integrates high-fidelity electrophysiology simulations with machine-learning–based emulators to assess drug-induced QT prolongation in a sex-specific manner.

Sex-specific virtual populations were generated using 3D finite-element cardiac electrophysiology models [1], simulating drug effects via a multi-channel pore-block model across key ion currents. From these simulations, pseudo-ECGs were extracted to quantify QT changes. To enable rapid risk evaluation, we developed Gaussian Process Regression emulators trained on over 900 3D simulations [3]. These emulators allow real-time predictions of QT prolongation with uncertainty quantification, achieving mean absolute errors below 4 ms.

As a proof of concept, we applied this framework to loperamide, a drug associated with abuse-related cardiotoxicity. The emulators were used to explore a wide concentration range beyond therapeutic exposure, identifying thresholds of arrhythmic risk across male and female profiles. Figure 1 illustrates the relationship between total concentration and QT prolongation (ΔQT), highlighting sex-specific risk thresholds and arrhythmic outcomes.

This case study demonstrates how AI-driven emulators can extend the reach of mechanistic models to high-throughput safety assessment, even in scenarios that would be unethical or infeasible to test clinically. This framework supports more efficient and comprehensive drug safety evaluations.

Predicted ΔQT under loperamide effect