Speaker
Description
Agent-based models (ABMs) have become valuable tools for understanding complex systems in biology and medicine. In order to evaluate the robustness and accuracy of the model predictions, uncertainty quantification using global sensitivity analysis should be performed. Unfortunately, most global sensitivity analyses are computational prohibitive for complex ABMs. By leveraging explicitly formulated surrogate models, we develop an efficient and flexible framework for inferring global sensitivity. We evaluate our methodology on two AMBs, one a simple 2D in vitro cell proliferation assay model and a second, more complex ABM of 3D vascular tumor growth. In these models, we simulate cells in their environment as decision making “agents” that have their own individual properties and interactions between agents, both temporally and spatially. In this talk, we show that our method for uncertainty quantification is comparable with other methods in accuracy, such as Morris one-at-a-time method or eFAST, but substantially speeds up the time it takes to complete the analysis from days to minutes. In addition, our method is able to estimate sensitives of ABM parameters that are not explicitly included in the surrogate model.
