Low-fidelity simulator selection for multi-fidelity inference of epithelial vertex models

PRISM screening shows that for a proliferating epithelial vertex model, early stopping in the low-fidelity simulator is unreliable and can yield confidently wrong parameter estimates after training.

• Why it matters: Bayesian parameter inference for stochastic proliferating tissues is expensive because it needs many full simulations, and it is unclear which low-fidelity approximations are safe to train on.
• What they did: The authors introduce PRISM, which ranks candidate low-fidelity simulators by paired-run agreement in cell-packing statistics versus cost, then train multi-fidelity inference using the screened choice.
• The result: With PRISM-selected low-fidelity settings, multi-fidelity inference outperforms approximate Bayesian computation and neural-network inference trained only on full simulations at matched cost with well-calibrated uncertainty, and a similar pattern holds for a Drosophila pupal-wing dataset where stopping halfway is safe.

AI summary of the abstract · may contain errors · not medical advice · read the paper.