A deep learning platform to assess drug proarrhythmia risk

Area of Science:

• Cardiology
• Pharmacology
• Biotechnology

Background:

• Human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) are used for in vitro drug safety testing.
• Predicting clinical arrhythmia risk from in vitro models remains a challenge.

Purpose of the Study:

• To develop a deep learning model for predicting drug-induced cardiac arrhythmia.
• To assess the correlation between in vitro arrhythmia features and clinical risk.
• To investigate the impact of patient genetics on drug-induced arrhythmia propensity.

Main Methods:

• A convolutional neural network (CNN) classifier was trained on hiPSC-CM action potential recordings.
• The CNN identified features associated with lethal Torsade de Pointes arrhythmia.
• hiPSC-CMs from healthy donors and patients with arrhythmogenic cardiomyopathies were used.

Main Results:

• The CNN accurately predicted clinical drug-induced arrhythmia risk.
• Drug risk profiles were consistent across hiPSC-CMs from different healthy donors.
• Pathogenic mutations exacerbated proarrhythmic responses to certain drugs.

Conclusions:

• Deep learning effectively identifies in vitro arrhythmic features predictive of clinical outcomes.
• hiPSC-CM models can discern genetic influences on drug-induced arrhythmia risk.
• This approach enhances the reliability of in vitro drug safety evaluations.