An arXiv preprint that compared two data-driven approaches for modeling calcium signaling in cell cultures has been withdrawn by the authors. The paper, posted in May 2025 and withdrawn in October 2026, reported that a Constrained Regularized Least-Squares Method (CRLSM) produced good parameter estimates and data fit for the governing ordinary differential equations, while Physics-Informed Neural Networks (PINNs) failed to match that performance under the tested configurations.
The authors had hypothesized PINNs would do well, but the results did not support that. They noted that only a limited number of PINN architectures were studied and suggested that hyperparameter tuning and uncertainty quantification could improve performance in future work.
Because the paper has been withdrawn, the findings should not be treated as established. The withdrawal note says a new model revealed issues with the results, so the reported comparison is unreliable as it stands.