ADMET-AI Enables Interpretable Predictions of Drug-Induced Cardiotoxicity.

Drug-induced cardiotoxicity (DICT) is a severe adverse drug reaction that affects the cardiovascular system and is a leading cause of drug withdrawals and clinical trial failures. DICT prediction is challenging because of the complex nature of cardiotoxicity, which could arise from a myriad of molecular pathways that lead to arrhythmias and cardiomyopathies, subsequently resulting in heart failure and sudden cardiac death. Experimental approaches using cardiac cells and animal models can be slow and expensive, and their results do not always correlate with DICT in humans. However, machine learning methods can be trained on real-world clinical DICT data to predict the cardiotoxic properties of drugs rapidly and accurately, thereby saving time and money by avoiding late-stage drug failure. Our machine learning platform ADMET-AI (admet.ai.greenstonebio.com), the fastest and most accurate publicly available ADMET web server, uses graph neural network models to predict absorption, distribution, metabolism, excretion, and toxicity (ADMET) properties of molecules. Because cardiotoxicity is influenced by multiple ADMET properties, this study extended ADMET-AI to predict DICT in an interpretable manner and shed light on the possible sources of DICT.