All new drug candidates must be tested for their potential to cause arrhythmia as a drug-induced
adverse event. Recent years have seen substantial progress in developing more sensitive and specific
predictions of which drugs may increase arrhythmia risk, and my group has been at the forefront of efforts to
employ mechanistic modeling for quantitative predictions of cardiac drug safety. Nonetheless, arrhythmias are
rare events, and even drugs that are considered dangerous only induce arrhythmias in a minority of patients.
Therefore, identifying which patients are most at risk of arrhythmia, and which conditions increase their risk,
is as important as classifying drugs. The work proposed here will address these challenging questions through
an innovative combination of: (1) in vitro physiology experiments that will quantify how drugs influence myocyte
action potentials and intracellular calcium; (2) simulations with mechanistic mathematical models that
incorporate phenotypic differences between groups and between individuals within the same group; and (3)
machine learning to synthesize results and develop predictive classification systems. Experiments performed
in stem cell-derived myocytes will measure cellular responses to a wide range of drugs, and these data will
allow for rigorous tuning of mathematical models. Subsequent simulations of heterogeneous populations will
address challenging unresolved questions, such as:
1. How do patient characteristics influence arrhythmia risk? Simulations will address how sex differences
in cardiac electrophysiology and the presence of pre-existing cardiac disease influence drug responses.
Through a combination of cellular experiments, mechanistic mathematical modeling of heart cells, and
machine learning models, we will quantify how much each factor influences arrhythmia risk.
2. How do symptoms associated with common diseases influence the potential pro-arrhythmic effects
of drugs used to treat those diseases? Many diseases are associated with conditions that influence cardiac
electrophysiology, such as fever, hypokalemia, and chronic inflammation. We will develop a simulation platform
that accounts for these effects.
3. Which patients within a group are especially at risk? Besides quantifying the effects of differences
between groups, our machine learning classifiers will allow us to predict which patients within a group are
especially susceptible to drug-induced arrhythmia on the basis of their “electrophysiological signatures.”
Together these studies will offer a new paradigm for quantitative understanding and prediction of drug-induced
arrhythmia that considers not only differences between drugs, but also between the patients that take these