In Silico Classifiers for the Assessment of Drug Proarrhythmicity

Drug-induced torsade de pointes (TdP) is a life-threatening ventricular arrhythmia responsible for the withdrawal of many drugs from the market. Although currently used TdP risk-assessment methods are effective, they are expensive and prone to produce false positives. In recent years, in silico cardiac simulations have proven to be a valuable tool for the prediction of drug effects. The objective of this work is to evaluate different biomarkers of drug-induced proarrhythmic risk and to develop an in silico risk classifier. Cellular simulations were performed using a modified version of the O'Hara et al. ventricular action potential model and existing pharmacological data (IC 50 and effective free therapeutic plasma concentration, EFTPC) for 109 drugs of known torsadogenic risk (51 positive). For each compound, four biomarkers were tested: T x (drug concentration leading to a 10% prolongation of the action potential over the EFTPC), T qNe (net charge carried by ionic currents when exposed to 10 times the EFTPC with respect to the net charge in control), T riang (triangulation for a drug concentration of 10 times the EFTPC over triangulation in control), and T EAD (drug concentration originating early afterdepolarizations over EFTPC). Receiver operating characteristic (ROC) curves were built for each biomarker to evaluate their individual predictive quality. At the optimal cutoff point, accuracies for T x , T qNe , T riang , and T EAD were 89.9, 91.7, 90.8, and 78.9% respectively. The resulting accuracy of the hERG IC 50 est (current biomarker) was 78.9%. When combining T x , T qNe and T riang into a classifier based on decision trees, the prediction improves, achieving an accuracy of 94.5%. The sensitivity analysis revealed that most of the effects on the action potential are mainly due to changes in I Kr , I CaL , I NaL and I Ks . In fact, considering that drugs affect only these four currents, TdP risk classification can be as accurate as when considering effects on the seven main currents proposed by the CiPA initiative. Finally, we built a ready-to-use tool (based on more than 450 000 simulations), which can be used to quickly assess the proarrhythmic risk of a compound. In conclusion, our in silico ool can be useful for the preclinical assessment of TdP-risk and to reduce costs related with new drug development. The TdP risk-assessment tool and the software used in this work are available at https://riunet.upv.es/handle/10251/136919.