BeatProfiler: Multimodal In Vitro Analysis of Cardiac Function Enables Machine Learning Classification of Diseases...

Area of Science:

• Cardiovascular Biology
• Biophysics
• Computational Biology

Background:

• Cardiac contractile function and calcium handling are crucial for understanding heart physiology and disease.
• Existing analytical methods for these metrics are often slow, error-prone, or require specialized resources.
• There is a need for efficient and accurate tools for in vitro cardiac model analysis.

Purpose of the Study:

• To develop and validate BeatProfiler, a suite of tools for quantifying cardiac contractile function and calcium handling in vitro.
• To apply machine learning for deep phenotyping and classification of cardiac disease models and drug responses.
• To improve the speed, accuracy, and robustness of cardiac function analysis.

Main Methods:

• BeatProfiler was developed to analyze contractile and calcium signals from various in vitro cardiac models.
• Benchmarking was performed against existing tools to assess accuracy, robustness, and speed.
• Machine learning models, including feature-based ML and TCN-BiLSTM, were employed for classification tasks.
• Grad-CAM was used to visualize drug perturbation signatures in calcium signals.

Main Results:

• BeatProfiler demonstrated a 7 to 50-fold increase in analysis speed compared to existing methods.
• It showed improved sensitivity and reduced false positives in signal detection.
• BeatProfiler accurately classified restrictive cardiomyopathy models with 98% accuracy and cardiac drugs with 96% accuracy.
• Key features distinguishing disease states and drug effects were identified.

Conclusions:

• BeatProfiler significantly enhances the efficiency and accuracy of in vitro cardiac studies.
• The tool facilitates rapid phenotyping and objective classification of cardiac diseases and drug responses.
• BeatProfiler is expected to advance research into cardiac health, disease mechanisms, and therapeutic interventions.