Classifying Drugs by their Arrhythmogenic Risk Using Machine Learning

Abstract

All medications have adverse effects. Among the most serious of these are cardiac arrhythmias. Current paradigms for drug safety evaluation are costly, lengthy, conservative, and impede efficient drug development. Here, we combine multiscale experiment and simulation, high-performance computing, and machine learning to create a risk estimator to stratify new and existing drugs according to their proarrhythmic potential. We capitalize on recent developments in machine learning and integrate information across 10 orders of magnitude in space and time to provide a holistic picture of the effects of drugs, either individually or in combination with other drugs. We show, both experimentally and computationally, that drug-induced arrhythmias are dominated by the interplay between two currents with opposing effects: the rapid delayed rectifier potassium current and the L-type calcium current. Using Gaussian process classification, we create a classifier that stratifies drugs into safe and arrhythmic domains for any combinations of these two currents. We demonstrate that our classifier correctly identifies the risk categories of 22 common drugs exclusively on the basis of their concentrations at 50% current block. Our new risk assessment tool explains under which conditions blocking the L-type calcium current can delay or even entirely suppress arrhythmogenic events. Using machine learning in drug safety evaluation can provide a more accurate and comprehensive mechanistic assessment of the proarrhythmic potential of new drugs. Our study paves the way toward establishing science-based criteria to accelerate drug development, design safer drugs, and reduce heart rhythm disorders.

Figure 1

Hybrid computational-experimental approach to quickly and reliably characterize the proarrhythmic potential of existing and new drugs. We characterize calcium transients in ventricular cardiomyocytes in response to drugs both computationally (top) and experimentally (bottom) and identify the ion channels that most likely generate early afterdepolarizations (left). We then screen the concentration space of the two most relevant channels and identify the classification boundary between the arrhythmic and nonarrhythmic domains using high-performance computing and machine learning (center). We validate our approach using electrocardiograms, both computationally and experimentally, in whole-heart simulations and isolated Langendorff perfused hearts (right). We demonstrate the potential of our new classifier by risk stratifying 23 common drugs and comparing the result against the reported risk categories of these compounds. To see this figure in color, go online.

Figure 2

Effect of different ion channels on the probability of early afterdepolarizations. Positive values imply that blocking this ion channel enhances early afterdepolarizations; negative values imply that blocking prevents early afterdepolarizations. Blocking the rapid delayed rectifier potassium current I_Kr and the L-type calcium current I_CaL has the strongest effect on enhancing and preventing early afterdepolarizations. To see this figure in color, go online.

Figure 3

Early afterdepolarizations in single-cell simulation and experiment. Isolated rat cardiomyocyte (top left) and the probability of developing early afterdepolarizations in response to the drug dofetilide at concentrations of 4, 8, 16, 38, and 130 nM (n = 6 cells each; top right) are shown. Calcium transients in response to the drug dofetilide at 0, 16, and 130 nM in the computational simulation (bottom left) and experiment (bottom right) are shown. To see this figure in color, go online.

Figure 4

Proarrhythmic risk classification. Screening the parameter space of rapid delayed rectifier potassium current I_Kr and the L-type calcium current I_CaL block reveals the classification boundary beyond which arrhythmias spontaneously develop. Blue electrocardiograms associated with the blue region displayed normal sinus rhythm; red electrocardiograms associated with the red regions spontaneously developed an episode of torsades de pointes. To see this figure in color, go online.

Figure 5

Ventricular arrhythmias in whole-heart simulation and Langendorff perfused hearts. Preparation of isolated rat heart (top left), four drug concentrations visualized in the proarrhythmic risk classification estimator (top middle), and risk of premature ventricular contractions and arrhythmias in response to varying concentrations of drugs dofetilide and nifedipine (n ≥ 6, ^∗p < 0.05 compared to (1), #p < 0.05 compared to (2); top right) are shown. Dofetilide selectively blocks the rapid delayed rectifier potassium current I_Kr; nifedipine selectively blocks the L-type calcium current I_CaL. Electrocardiograms in response to dofetilide at 0 and 20 nM combined with nifedipine at 0, 60, and 480 nM in the computational simulation (bottom left) and experiment (bottom right) are shown. To see this figure in color, go online.

Figure 6

Risk stratification of 23 drugs using our proarrythmic risk classification. Black and white regions indicate fibrillating and nonfibrillating regimes; red and blue curves represent the I_Kr/I_CaL profiles of high- and low-risk drugs at varying concentrations; gray dots and numbers indicate the critical concentration at which the drug curves cross the classification boundary as predicted by our proarrhythmic risk classification in Fig. 4. For comparison, numbers from 1 to 5 indicate the reported torsadogenic risk (20); red and blue colors of the numbers indicate torsadogenic and nontorsadogenic compounds (19). To see this figure in color, go online.

Figure 7

Experimental validation of risk stratification for 12 drugs. Black lines represent the experimentally measured critical concentration in isolated rabbit hearts (42, 43, 44). Stars indicate classification by early afterdepolarization, torsades de pointes, and ventricular tachycardia. Dots represent our predicted critical concentration, with red indicating proarrhythmic and blue safe drugs (20). To see this figure in color, go online.

Figure 8

Computational validation of risk stratification for three drugs applied at the same concentration. At 10× the effective free therapeutic concentration, terfenadine blocks 84% of I_Kr and 11% of I_CaL, bepidril blocks 86% of I_Kr and 50% of I_CaL, and verapamil blocks 79% of I_Kr and 84% of I_CaL. The different degrees of blockade result in arrhythmic patterns for terfenadine and bepidril, but not for verapamil, for which the high degree of I_CaL block prevents the development of arrhythmia and slows the beating rate. To see this figure in color, go online.