Speaker
Description
Human induced pluripotent stem cell-derived cardiomyocytes (hiPSC--CMs) provide a physiologically relevant platform for studying cardiac electrophysiology under both healthy and pathological conditions, as well as for drug cardiotoxicity screening. Multi-electrode arrays (MEAs) enable non-invasive, long-term recording of extracellular field potentials from hiPSC-CM monolayers, capturing cellular electrical activity across pharmacological perturbations and disease phenotypes.
Existing ionic models for hiPSC-CMs — for both ventricular--like and atrial--like phenotypes— replicate action potential morphology by simulating voltage-dependent ion channel kinetics and intracellular calcium handling. While these models capture the underlying biophysics, their computational cost becomes prohibitive when exploring large parameter spaces, performing uncertainty quantification, or solving inverse problems to estimate conductances from experimental data. Approaches through population of models, which generate heterogeneous virtual cohorts by sampling conductance parameters, further exacerbate this computational burden.
We present a deep learning framework that leverages neural network emulators to approximate hiPSC-CM ionic models with several orders of magnitude speedup while maintaining sub-millisecond accuracy. Trained on synthetic populations, the emulator enables real-time forward simulation across physiological variability.
This methodology bridges high-throughput experimental electrophysiology with mechanistic modeling, providing a scalable tool for drug safety assessment, disease phenotyping, and precision cardiology. By reducing simulation time, our approach unlocks previously intractable applications in parameter estimation, sensitivity analysis, and virtual clinical trials.