BPS2026 – Reconstructing cardiac dynamics and model parameters using data assimilation

Cardiac ventricular action potentials arise from nonlinear interactions between transmembrane voltage, ionic currents, and intracellular ion concentrations. Current experimental methods cannot simultaneously record membrane potential and multiple ionic currents, limiting insight into how perturbations, such as altered pacing or drug treatment, initiate or prevent arrhythmias. Biophysical cell models can simulate these dynamics, but their parameters are often averaged or extrapolated across experiments, introducing uncertainty and reducing their ability to capture biological heterogeneity. We address this challenge using data assimilation, a computational framework that fuses mathematical models with noisy observations to jointly estimate state variables and parameters. A heterogeneous virtual population of guinea pig ventricular cells was generated by systematically varying ionic conductances, with Gaussian noise added to simulated transmembrane potential and intracellular calcium recordings. Our approach reconstructed the full state space, including membrane potential, with error below the noise level, and accurately estimated ionic conductances. Compared to two standard post hoc optimization methods, data assimilation achieved higher accuracy with shorter computational time. As a proof of concept, we applied the method to a virtual population of human iPSC-derived ventricular cells, successfully reconstructing complex action potential morphologies and critical parameters. These results demonstrate that data assimilation enables precise state reconstruction and parameter estimation in heterogeneous populations, advancing the use of biophysical models for mechanistic studies, drug response prediction, and interpretation of noisy electrophysiological data.