A deep learning algorithm to translate and classify cardiac electrophysiology.

Area of Science:

• Cardiology
• Stem Cell Biology
• Computational Biology

Background:

• Induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs) are valuable for patient-specific disease modeling but suffer from low throughput and immaturity.
• Existing methods for analyzing iPSC-CMs are limited in speed and phenotypic representation.
• There is a need for advanced computational tools to overcome the limitations of iPSC-CM platforms.

Purpose of the Study:

• To develop a deep learning multitask network for analyzing iPSC-CMs.
• To address challenges of low throughput, high variability, and immature phenotypes in iPSC-CM research.
• To enable accurate prediction of electrophysiological perturbations and drug effects in iPSC-CMs.

Main Methods:

• A novel deep learning multitask network was designed, combining translation and classification tasks.
• The network was trained using simulated action potential (AP) data.
• The model was applied to classify cells and predict electrophysiological changes across different maturation stages.

Main Results:

• The deep learning network successfully classified cells into drug-free and drugged categories.
• The network accurately predicted the impact of electrophysiological perturbations.
• Key information for multitasking was identified in the AP phase sensitive to perturbation.
• The network demonstrated successful translation of both experimental and simulated iPSC-CM AP data.

Conclusions:

• The developed deep learning network offers a powerful tool for analyzing iPSC-CMs.
• This approach can significantly improve the throughput and accuracy of drug screening and disease modeling.
• The study validates the network's ability to predict drug-induced effects on cardiomyocytes.