Machine Learning Prediction of Pre-Clinical Drug Induced QT Prolongation Liability and Pathway Analysis

Marked drug-induced prolongation of the QT interval on the electrocardiogram is associated with Torsades de Pointes, a potentially life-threatening cardiac arrhythmia. QT prolongation is also associated with a number of clinical factors including but not limited to genetic mutations, female gender, diurnal variability, ion concentration imbalance, pathological conditions, excessive drug exposure, or a combination of these. While QT prolongation is known to be an imperfect safety biomarker, it is currently a widely accepted standard. Assessment of QT prolongation liability in the drug development process is required but is time and resource intensive. Current pre-clinical safety assessment paradigms include patch clamp analysis to evaluate drug-related block of cardiac repolarizing currents. Previously limited to the Human Ether-a-Go-Go (hERG) potassium rectifier current channel, analyses have broadened with the Comprehensive in vitro Proarrhythmia Assay (CiPA) initiative to include patch clamp analysis of other ion channels and the use of in silico models of cardiac ion channels. Recently reported hERG patch clamp sensitivities and specificities range from between 64-82% and 75-88% respectively. SVM classifier results in this project using 10-fold cross validation of rat liver RNA expression profiles to predict human QT prolongation liability have shown average sensitivities of 85% and specificity of 90% using all available datasets as input, and a mean sensitivity and specificity of 92% and 94%, respectively across 77 drug sub-classifications. Clustering based on expression profile similarities in each of the 77 classifications showed that while drugs known to prolong QT interval do not always cluster into “pure” groups, the number of groups was limited. An attempt was made to link physiological phenomena using genes associated with congenital Long QT Syndrome, genes associated with autonomic activity, and drug-induced gene expression changes using the most significantly differentially expressed probes in each of the 77 classes. The most common network connections involved changes in fatty acid metabolism, associations with G proteins, associations with glutathione, immune responses and apoptosis, mitochondrial activity and electron transport, and mitogen activated protein kinases. These results suggest that machine learning methods may augment cardiac safety predictions during drug safety assessments.