ABSTRACT
Multiscale computational models of heart are being extensively investigated for improved assessment of drug-induced Torsades de Pointes (TdP) risk, a fatal side effect of many drugs. Model-derived metrics (features) such as action potential duration and net charge carried by ionic currents (qNet) have been proposed as potential candidates for TdP risk stratification after being tested on small datasets. Unlike purely statistical approaches, model-derived metrics are thought to provide mechanism-based classification. In particular, the underlying mechanism behind the success of the recently proposed qNet metric is attributed to its correlation to early afterdepolarizations (EADs), which are known to be cellular triggers of TdP. Analysis of critical model components and of ion-channels that have major impact on model-derived metrics can lead to improvement in the confidence of the prediction. In this paper, we analyze a large population of virtual drugs to systematically examine the influence of different ion channels on model-derived metrics that have been proposed for proarrhythmic risk assessment. Global sensitivity analysis (GSA) methods were employed to determine and highlight the critical input parameters that affect different model-derived metrics. We observed significant differences between the sets of input parameters that control model-derived metrics and generation of EADs in the model, thus opposing the idea that these metrics and sensitivity to EAD might be strongly correlated. Moreover, in classification of a small set of actual drugs, we found that the classifiers based on EADs performed worse than those built on other model-derived metrics. Hence, our analysis points towards a need for a better mechanistic interpretation of promising metrics such as qNet based on formal analyses of models. In particular, GSA should constitute an essential component in the in silico workflow for proarrhythmic risk assessment to yield improved understanding of the structure of mechanistic dependencies surrounding model-derived metrics while ultimately providing increased confidence in model-predicted risk.