Cardiac muscle contraction is governed by the cross-bridge cycle, where myosin heads cyclically attach to actin filaments to generate force. Accurate modeling of this process is essential for understanding sarcomere mechanics and muscle dysfunction. In this study, we developed deep learning models using artificial neural networks (ANNs) to simulate cardiac muscle contraction under isosarcometric, isometric, and isotonic conditions. Trained on synthetic data, the ANNs captured nonlinear relationships among calcium concentration, sarcomere length, temperature, and muscle force generation. Error analysis with histograms and unity-line scatter plots validated the model prediction accuracy. Comparisons across network architectures showed the effect of hidden layer complexity on generalization. The models reproduced a wide range of physiological behaviors, including force-Ca 2+ relations, force-velocity profiles, and sarcomere length changes, with predictions closely matching theoretical benchmarks. Importantly, the deep learning model can also predict cardiac twitch dynamics as a function of sarcomere length and calcium activation, and cell shortening as a function of Ca 2+ , where the cell shortens from rest length against a passive restoring force. These findings demonstrate the potential of deep neural networks to advance cardiac muscle modeling and enable novel in-silico methods for predicting cardiac disease mechanisms, such as cardiomyopathies.
Yasser Aboelkassem (Sun,) studied this question.