Computational model credibility

Inverse Uncertainty Quantification of RBC mechanics in HemoCell

Problem

Cell-resolved blood models depend on membrane-parameter choices that are expensive to calibrate directly with full simulations. I addressed how to estimate parameter uncertainty and identifiability without prohibitive compute cost.

Approach

  • Formulate inverse uncertainty quantification with Bayesian Annealed Sequential Importance Sampling.
  • Train a Gaussian process surrogate to emulate expensive simulation outputs.
  • Use Sobol indices to assess parameter identifiability.
  • Calibrate against shear-driven RBC elongation behavior.

Key finding

The surrogate-accelerated pipeline recovered identifiable parameter groups efficiently and improved agreement with measurements compared with baseline parameter sets.

Why it matters

Credible uncertainty-aware calibration strengthens trust in predictive simulation results and clarifies which parameters need better experimental constraints.

Outputs

  • Publication details are listed in the References section below.
  • Calibration and identifiability workflow summary is presented on this page.

References

2020

  1. Inverse uncertainty quantification of a cell model using a gaussian process metamodel
    Kevin Vries, Anna Nikishova, Benjamin Czaja, and 2 more authors
    International Journal for Uncertainty Quantification, 2020