The rhythm of life arguably starts with our heartbeat—whose constant pulsing changes tempo in response to our daily activities and emotional state. For people living with cardiac arrhythmias, or irregular heartbeats, this steady beat can change unpredictably with potentially dangerous consequences.
Despite being a leading cause of heart failure, cardiac arrhythmia has remained notoriously difficult to diagnose and prevent. However, a stem-cell breakthrough may offer some hope: skin cells from a patient can be chemically reverted to an embryonic stem cell state before being coaxed into heart cells. This ‘heart in a dish’ allows researchers to model heart physiology in a relatively non-invasive manner, and without the need for human embryos to be harvested.
The next hurdle is matching heart cell (cardiomyocyte) behaviour with disease risk, said Boon Seng Soh, a Principal Investigator at A*STAR’s Institute of Molecular and Cell Biology (IMCB). “While some in vitro phenotypes have been linked with arrhythmia, we still lack a robust method for identifying key electrophysiological changes associated with heart conditions,” explained Soh.
Working with a team from the National University of Singapore, Soh and colleagues harnessed the power of machine learning (ML) algorithms to help bridge this gap and analyse large calcium-cycling datasets, a lab test used to assess the heart’s electrical activity.
The team first collected training data from human pluripotent stem cell-derived cardiomyocytes (hPSC-CMs) with a special reporter to visualise how calcium moves through the heart cell during contraction. They then input over 3,000 calcium-cycling data points collected from healthy and arrhythmic cardiomyocytes. Going a step further, Jeremy Pang, the leading author of the study, also trained binary classifiers to distinguish between specific subtypes of arrhythmia.
In their study, the scientists found that their ML algorithm could predict the presence (and subtype) of arrhythmia with over 90 percent accuracy. Soh said this predictive power could empower cardiologists to pick up on the condition's early signs in patients long before symptoms manifest. “Using cardiomyocytes generated from human induced pluripotent stem cells, we potentially can predict the types of arrhythmias the person may develop years later.”
Additionally, ML-driven platforms could help accelerate the screening of medications for any potential toxic effects on the heart. Personalised testing of drugs on patient-derived cardiomyocytes could help avoid heart-related side effects.
Soh, the study’s corresponding author, said the team plans to tweak the ML platform to handle data from miniaturised heart tissue culture models. “We intend to adapt the platform to work with 3D-chambered cardiac organoid in vitro models which we have recently developed,” Soh said, adding that they also hope to apply ML to tackle age-related heart failure.