Parameter uncertainty quantification using surrogate models applied to a spatial model of yeast mating polarization.

Area of Science:

• Systems Biology
• Computational Biology
• Uncertainty Quantification

Background:

• Estimating parameters and quantifying effects of unknown parameters in biological models is computationally challenging.
• Expensive model evaluations and numerous parameters often render these tasks intractable.

Purpose of the Study:

• To develop and demonstrate a computationally efficient method for sensitivity analysis and parameter estimation in complex biological models.
• To improve computational efficiency by replacing direct model simulations with polynomial surrogate model evaluations.

Main Methods:

• Utilized polynomial surrogate models to approximate complex biological models, reducing computational cost.
• Employed Markov chain Monte Carlo (MCMC) with surrogate models for rapid Bayesian inference of parameters.
• Applied the method to ODE and spatial PDE models of mating in budding yeast.

Main Results:

• Achieved significant computational efficiency gains by using surrogate models instead of direct simulations.
• Successfully performed global parameter sensitivity analysis and Bayesian inference for models with numerous parameters.
• Identified parameter distributions and correlations, revealing insights into yeast mating and polarization.

Conclusions:

• The developed surrogate model approach offers a computationally tractable solution for analyzing complex biological systems.
• This method enables efficient parameter estimation and sensitivity analysis, even for large-scale models previously considered prohibitive.
• Findings suggest distinct diffusion constants for active and inactive Cdc42, aligning with experimental observations.