Uncertainty Quantification in Complex Models of Complicated Biology

Validated mathematical models of complex biological phenomena are increasingly recognized as invaluable for elucidating mechanisms that underlie real-world (experimental or clinical) observations. Agent-based models (ABMs) have emerged as a natural formulation of choice in such models, providing a logical structure for capturing the multiple time and spatial scales associated with complex biology. However, the inherent stochasticity and heavy computational requirements of ABMs are significant obstacles for data-driven parameterization and for conducting parameter space exploration and sensitivity analyses. Further, experimental or clinical data used for parameterization is typically limited, coarse-grained and may lack spatial resolution, resulting in issues of parameter identifiability. There is hence a need for developing new theoretical and computational frameworks that can bridge the current divide between ABM parameters and real-world data. In this talk I will present a novel approach — Surrogate Modeling for Reconstructing Parameter Surfaces (SMoRe ParS) — that uses explicitly formulated surrogate models (SMs) to bridge ABM simulations and experimental data. Our approach quantifies the relationship between parameter values across the ABM and SM, and between SM parameters and experimental data. Thus, SM parameters act as interlocutors between ABM inputs and data that can be used for calibration and uncertainty quantification of ABM parameters. Time permitting, I will also present an extension of this work — Surrogate Modeling for Recapitulating Global Sensitivity (SMoRe GloS) — for performing global sensitivity analysis on ABM parameters.