Machine Learning Techniques to Classify Healthy and Diseased Cardiomyocytes by Contractility Profile

Area of Science:

• Biomedical Engineering
• Cardiology
• Computational Biology

Background:

• Human induced pluripotent stem (iPS) cell-derived cardiomyocytes are vital for studying cardiac function and drug testing.
• Machine learning (ML) shows promise in analyzing biological signals, including calcium transients from cardiomyocytes.
• Existing methods for cardiomyocyte analysis often involve terminal assays, limiting longitudinal studies.

Purpose of the Study:

• To develop and validate a machine learning approach for identifying healthy and diseased cardiomyocytes using noncontact, label-free contractility profiles from brightfield videos.
• To assess the utility of this ML method in distinguishing cardiomyocytes from patients with Timothy Syndrome (TS) and healthy controls.
• To explore the potential of this approach for broader applications in cardiac research and drug discovery.

Main Methods:

• Acquired brightfield videos of cardiomyocytes derived from human iPS cells, including those from TS patients and healthy individuals.
• Processed video data to extract contractility profiles.
• Developed and applied machine learning algorithms to classify cardiomyocytes based on their contractility profiles.

Main Results:

• The ML algorithms successfully distinguished cardiomyocytes from TS patients (a long QT disease) from healthy controls.
• The algorithms could also classify two different healthy control groups.
• The contractility-based ML approach proved effective as a noncontact, label-free method.

Conclusions:

• Computational ML evaluation of iPS cell-derived cardiomyocyte contractility offers a powerful tool for disease identification and characterization.
• This method facilitates longitudinal studies and can be extended to organs-on-chip models.
• Potential applications include identifying genetic proarrhythmic events, screening therapeutic agents, and predicting drug-target interactions.