Drug development and approval is a costly process with nearly $2 billion spent for each drug that is approved. One of the most significant contributors to this high cost is the expense of developing drugs that fail to pass clinical trials ? only 15% of drugs that begin clinical trials are approved for use on humans. A common reason for not reaching the FDA?s criteria for approval is that the drug is classified as cardiotoxic, which cannot be detected during early stages of drug development. One way that drugs can lead to cardiotoxicity is by altering the electrical activity of ion channels that are responsible for the excitation of the heart tissue that pumps blood to the body. Understanding which ion channels, and the extent to which these ion channels are affected is central to determining the cardiotoxicity of a drug. Early-stage predictions of cardiotoxicity are based on animal studies that are poor models of human heart behavior or single-cell electrophysiological studies that falsely assume underlying pathophysiology based on action potential changes. The recent development of human induced pluripotent stem cell-derived cardiomyocytes (hiPSC-CMs) offers an opportunity to study drug effects on human cells in a preclinical setting. In this study, we hypothesize that fitting a computational iPSC-CM model to voltage-clamp (VC) data acquired before and after drug application provides a means of quantifying unknown drug effects on specific cardiac ion channels. We will address this hypothesis through the following Specific Aims: 1) Use machine learning to design a novel VC protocol that improves the quality of data for hiPSC-CM model fitting. 2) Quantify the change in hiPSC-CM ion channel conductances before and after drug application. We will use machine learning to develop a novel voltage clamp protocol that improves the electrophysiology data acquired from our hiPSC-CMs. The data is optimized to improve predictions of the conductances for all ion channels activated during the cardiac action potential. We can apply this voltage clamp protocol in an in vitro setting before and after drug application, then fit our computational model to each dataset. The change in ion channel conductances, predicted by the model fit, serves as an estimate of the channel-specific effects of the drug. The contributions of this proposal will be significant because it will be the first study to use human cardiac cells to produce quantitative measurements of channel-specific drug targets that may lead to lethal cardiac arrhythmias.
Many promising drugs developed to treat a variety of diseases have undesirable effects on the heart, including increased risk of sudden cardiac death. Often, these drugs are not found to have adverse effects until they reach clinical trials, and failure of drugs at this stage of development is very costly. We aim to develop a method that leverages machine learning and human stem cell-derived heart cells to predict drug cardiotoxicity in a pre-clinical setting to substantially reduce the cost of the drug development process.
