eMB: Bridging the Gap Between Agent Based Models of Complex Biological Phenomena and Real-World Data Using Surrogate Models

Much has happened in the past three years - to us as individuals, as a united nation, and as one world. The consequences of human-induced transformations on our environment and the reciprocal impact of the changing global environment on humanity have been profound. To address these challenges and improve human well-being, researchers, scientists, and engineers are generating large amounts of data on the evolving condition of our world and its inhabitants in novel, multidimensional forms.

Unfortunately, existing mathematical, statistical, and computational techniques offer only partial tractability in comprehending these complex datasets. New, thoughtfully developed mathematical methods and modeling approaches are desperately needed to gain a deep and robust understanding of these data for human benefit and to mitigate human harm. The successful completion of this project will result in a robust and scalable computational framework for constraining large parameter spaces in agent-based models with real-world data.

Agent-based models are widely recognized as a powerful computational framework for advancing our understanding of human disease, human-society interactions, and environmental systems. However, their inherent stochasticity and prohibitive computational expense pose significant barriers to integrating such models with real-world data. The new approach will provide a much-needed platform for exploring parameter uncertainty and sensitivity in multiscale agent-based models representing complex biological phenomena.

Ultimately, the new methods developed here will result in a scalable mathematical tool for operationalizing computationally complex models designed to solve formidable biological problems that are of great interest to biologists, ecologists, clinicians, and health policymakers.

Unlocking the full potential of computationally complex mathematical models to advance our understanding of interconnected biological systems urgently requires techniques for integrating these models with multifaceted real-world data. Multiscale agent-based models (ABMs) are widely recognized as a powerful computation framework for advancing our understanding of systems ranging from molecular, cellular, and tissue dynamics to human-society interactions, infectious diseases, and ecological systems.

However, to make meaningful, reliable quantitative predictions and to gain mechanistic insight, ABMs must be integrated with real-world data through model parameterization and calibration. Unfortunately, robust, scalable techniques for addressing the challenges posed by the inherent stochasticity and heavy computational requirements of an ABM in integrating it with real-world data are sorely lacking.

Hence, there is a critical need to develop new theoretical and computational approaches to bridge this gap between ABM parameters and real-world data. This project develops a new computational framework for parameter estimation, uncertainty quantification, and sensitivity analysis of multiscale ABMs informed by noisy, sparse, and multifaceted real-world data.

The method utilizes explicitly formulated and mechanistic surrogate models simultaneously inferred from both the ABM formulation and the data to link the two in previously impossible ways. The approach will open new possibilities for ABMs representing complex biological phenomena to uncover how data sets can hide unexpected or counter-intuitive underlying mechanisms that have profound implications for predicted outcomes and planned interventions.

This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.