Digital Twin for the Win: Personalized Cardiac Electrophysiology

Abstract

Background: Individual variability influences disease susceptibility, therapeutic response, and the emergence of rare phenotypes in both inherited and acquired cardiac diseases. Conventional preclinical models intentionally reduce variability to isolate biological effects but consequently fail to capture the heterogeneity present in human populations. This limitation contributes to translational gaps, incomplete mechanistic understanding, and adverse drug reactions. Digital twins offer a novel solution by integrating individualized data with simulation-based inference to predict physiology and therapeutic outcomes in a cell-specific, and ultimately, patient specific manner.

Methods: We developed an integrated computational, experimental, and machine learning framework to generate digital twins of human induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs). A synthetic population of over one million computational iPSC-CMs was created by introducing physiologically plausible variation in 52 biophysical parameters governing six major ionic currents. Two synthetic datasets were used to train and test a fully connected deep neural network to infer complete parameter sets directly from optimized whole-cell voltage-clamp recordings.

Results: Application of the model to experimental iPSC-CMs enabled rapid extraction of ionic conductance and kinetic parameters, generating digital twins that reproduced action potential waveforms and pacing frequencies. The models captured fine-scale electrophysiological features of depolarization, plateau, and repolarization with high fidelity across diverse morphologies and recording conditions. The framework was robust to both temperature perturbations and broad morphological variability, as we show that the synthetic training data can be readily re-tuned to any recording temperature and to include broad AP phenotypes.

Conclusions: This work introduces a scalable technology for generating fully parameterized, cell-specific digital twins of human iPSC-CMs from a single recording. By unifying computational modeling, synthetic data generation, and deep learning, the approach transforms a slow, multi-step process into a rapid, versatile platform for personalized diagnostics, targeted therapeutics, and predictive safety pharmacology.

Keywords: Artificial Intelligence; Cardiac Electrophysiology; Computational Biology; Digital twin; Precision Medicine; Stem Cell Biology; computational modeling and simulations; deep learning; iPSC-CM; personalized cardiac electrophysiology.

Figure 1:. The variability of APs in simulated iPSC-derived cardiomyocyte populations.

(A) To illustrate population variability, 20 APs were shown, each resulting from ±40% random variation applied to 52 parameters governing six key ionic currents (I_Kr, I_CaL, I_Na, I_Ks, I_K1, and I_f,) in the baseline Kernik iPSC-CM model, within a simulated population of 200,000 spontaneously beating cells. (B) These perturbations yielded a wide spectrum of APs, with substantial variation in both waveform and frequency. (C) The corresponding total ionic current and its decomposition into individual current components are visualized in panels (I_Kr, I_CaL, I_Na, I_Ks, I_K1, and I_f, D–I, respectively).

Figure 2.. Digital twins of human iPSC-CMs from one simple voltage-clamp recording.

Top row (pipeline overview): A large population of synthetic iPSC-CMs (top left panel) is generated by introducing variation to 52 biophysical parameters governing key ionic currents in the baseline model: I_Kr, I_CaL, I_Na, I_Ks, I_K1, and I_f (highlighted with red asterisks in the schematic on right). A computationally optimized voltage-clamp protocol (black trace, top left center panel) is applied to generate distinct whole cell current (I_Total, red trace), enabling cell-wide excitation of key ion channels. The simulated whole cell currents I_Total from 200,000 synthetic cells serve as inputs to a fully connected neural network, trained to map raw current responses to the underlying 52 model parameters with high accuracy, as demonstrated by the low MSE across training and test sets (top center right panel). The deep learning model is trained to predict optimized parameter values by inferring gating kinetics and maximal conductance for each ionic species. An example formulation for the fast sodium current (I_Na) is shown, with inferred parameters (x₁–x₅) contributing to gating and conductance (top right panel). In the bottom row, the resultant digital twin output is shown with the inferred parameters used to simulate the AP, whole cell currents I_Total and each key ion channel I_Na, I_CaL, I_K1, I_Ks, I_Kr, and I_f.

Figure 3:. Simulated training dataset for parameter extraction using an empirical voltage clamp protocol.

(A) The empirical membrane voltage protocol (Vm, top) applied to in silico iPSC-CMs. The bottom panel shows the resulting whole-cell current response (red), with an inset providing an enlarged view of the current trace for better visualization of transient events. The time course of the whole-cell current was used as input for a deep learning model to extract key ion channel parameters. (B–G) Simulated individual ionic currents generated during the same protocol: (B) L-type calcium current I_CaL, (C) rapid delayed rectifier potassium current I_Kr, (D) fast sodium current I_Na, (E) slow delayed rectifier potassium current I_Ks, (F) hyperpolarization-activated funny current I_f, and (G) inward rectifier potassium current I_K1. Each trace demonstrates the distinct temporal activation patterns and amplitudes of the respective ionic species under the applied voltage protocol.

Figure 4:. An empirical protocol for estimating ion channel parameters.

(A) A voltage-clamp protocol is applied to induce characteristic current responses (I_total red trace, right) in response to dynamic changes in membrane potential (V_m, black trace), effectively stimulating key ion channels across the cell. The inset highlights the transient components of the current response. (B) Prediction error for each model parameter was evaluated using training datasets of sizes 1,000, 10,000, and 200,000 (left to right). For smaller datasets, the test error exhibited an asymmetric U-shaped curve, while the training error consistently decreased over iterative cycle (Epoch). Mean absolute error (MAE) distributions for each dataset size are shown in the bottom panels, with median MAE values indicated. Increasing training data size reduced variability and improved agreement between training and test errors, indicating enhanced model generalization. (C) Simulated APs, intracellular calcium transients (Ca²⁺), whole-cell current, and six individual ionic currents; I_Kr, I_CaL, I_Na, I_Ks, I_K1, and I_f, generated using single-cell model parameters predicted by the deep learning (DL) network trained on 200,000 data points (lowest MAE case from panel B, right). Blue traces represent ground truth input data, and red traces represent deep learning model predictions, shown over a 1600 ms time window.

Figure 5:. Deep learning guided optimization of an ideal voltage clamp protocol.

Each iteration began with a −100 mV holding potential for 250 ms, followed by sequential testing potentials from −120 mV to +50 mV in 10 mV increments. A total of 200,000 synthetic samples were generated per training cycle. The optimal testing potential, identified by the lowest MSE from the deep learning model, was then applied for 250 ms. This optimization cycle repeated every 7000 ms. At 6000 ms, the potential was transiently stepped to −120 mV for 250 ms before resuming the next testing potential sequence. The schematic illustrates the iterative loop of model evaluation, MSE-based selection, and protocol updating, leading to the optimized composite voltage waveform shown in the lower panel.

Figure 6.. Deep learning for ion channel parameter estimation.

(A) A deep learning derived optimized voltage-clamp protocol (left panel) was designed to activate a broad set of ionic currents in iPSC-CMs by applying dynamic membrane potential steps (black trace), producing distinctive whole-cell current responses (I_Total, red trace). The inset highlights fine-scale current kinetics captured during the protocol. These V_m–I_Total pairs serve as network inputs for parameter inference. (B) Prediction accuracy for individual model parameters was evaluated using training datasets of 1,000, 10,000, and 200,000 synthetic cells (left to right). For the smallest dataset, the test error exhibited an asymmetric U-shaped curve across training epochs, indicating limited generalizability. As dataset size increased, training and test errors converged, with median mean absolute errors (MAE) of 0.041, 0.038, and 0.020 for the three dataset sizes, respectively. Bottom panels show MAE distributions across parameters, demonstrating progressively narrower error ranges with larger training datasets. (C) APs (V_m), intracellular calcium transients (Cai), total ionic current (I_Total), and six major individual ionic currents (I_Kr, I_CaL, I_Ks, I_K1, I_Na, I_f) simulated from the predicted parameters of a single test cell using the deep learning network trained on 200,000 samples. Blue traces represent the original simulated data (input), and red traces show the model outputs. The close overlap confirms accurate parameter recovery and faithful reproduction of cell-specific electrophysiological behavior.

Figure 7.. Digital twin generation and predictive modeling of iPSC-CM APs.

(A, B, top) Ten simulated APs from a synthetic population of 1,100,000 induced pluripotent stem cell-derived cardiomyocytes (iPSC-CMs) at room-temperature were generated by introducing ±40% random variation to 52 biophysical model parameters from baseline, governing six major ionic currents: (I_Kr, I_CaL, I_Ks, I_K1, I_Na, I_f). Colored traces represent exemplar simulated cells from training data; overlaid black traces are experimentally recorded APs from two representative iPSC-CMs from hiPSC cell line iPS-6-9-9T.B. Insets show the experimental whole-cell current responses to a deep learning optimized voltage-clamp protocol (as described in Figure 5), used for parameter inference. (C, D, bottom) Digital twin models of the same experimental iPSC-CMs shown in panels A and B. Digital twins were created by extracting all 52 model parameters from the experimental whole-cell current using the deep learning inference pipeline. These parameters were used to instantiate cell-specific computational models, and AP simulations (red traces) were generated. The close overlay between the experimental traces (black) and digital twin predictions (red) demonstrates the success of the framework to accurately derive cell-specific digital twins from real cells.