ABSTRACT: Drug-induced proarrhythmia is so tightly associated with prolongation of the QT interval that QT prolongation has become widely accepted as a surrogate marker for arrhythmia. The problem is that QT interval as an arrhythmia indicator is too sensitive and not selective, resulting in many potentially useful drugs eliminated early in the drug discovery process. We first set out to predict the fundamental mode of binding for the proarrhythmic drug dofetilide with the promiscuous cardiac drug target, the hERG potassium channel. In a novel linkage between the atomistic and functional scales, computed binding affinities and rates from atomistic simulation are utilized here to parameterize function scale kinetic models of dofetilide interactions with the hERG channel. The kinetic model components are then integrated into predictive models at the cell and tissue scales to expose fundamental arrhythmia vulnerability mechanisms and complex interactions underlying emergent behaviors. Human clinical data from published studies were used to validate model framework and showed excellent agreement, demonstrating feasibility of the approach. The model predictions show that a clinically relevant dose of dofetilide increased arrhythmia vulnerability in all emergent TRIaD-linked parameters including Triangulation, Reverse use-dependence, beat-to-beat Instability and temporal and spatial action potential duration Dispersion. Application of machine learning demonstrated redundancy in the TRIaD linked parameters and suggested that changes in beat-to-beat instability were highly predictive of arrhythmia vulnerability in this setting. Here, we demonstrate the development and validation of a prototype multiscale model framework to predict electro-toxicity in the heart for the proarrhythmic drug dofetilide from the atom to the rhythm.