Abstract
Microelectrode arrays (MEAs) provide powerful platforms for large-scale in vitro cardiac tissue analysis. We present a machine learning (ML)-assisted framework for multimodal characterization of cardiomyocyte activity recorded using a transparent thin-film-transistor MEA, allowing simultaneous acquisition of electrophysiological and optical data. The features of both modalities were integrated, and unsupervised learning identified contraction patterns without prior labels. Reliability was assessed with supervised models and SHapley Additive Explanations to confirm feature importance and label consistency. Pharmacological validation with isoprenaline showed the expected increase in beating frequency, detected automatically by our pipeline. Multimodal feature correlations with emergent labels demonstrated that interpretable ML can extract biologically meaningful insights from complex datasets. While isoprenaline served as a proof-of-concept, the framework is applicable for detecting diverse drug- or disease-induced cardiac network changes. This approach enables scalable, label-free and automated assessment of cardiomyocytes networks for drug discovery, toxicity screening, and mechanistic studies.
Funding Information
  • Murata Science Promotion and Education Foundation 2024 (M24AN156)
  • Institute of Industrial Science -UTokyo
  • Spring GX
  • Japan Science and Technology Agency Support (JPMJSP2108)
  • Japan Society for the Promotion of Science
  • Japan Agency for Medical Research and Development
  • Japan Science and Technology Agency (22gm6510010h0002)
  • Nanken-Kyoten, Science Tokyo (JPMXP1323015483)
  • Science Tokyo
  • Medical Research Center Initiative for High Depth Omics, Science Tokyo

This publication has 33 references indexed in Scilit: