Inference in epidemiological agent-based models using ensemble-based data assimilation.

Area of Science:

• Epidemiology
• Computational modeling
• Data assimilation

Background:

• Traditional compartmental models using ordinary differential equations capture mean-field disease dynamics but miss individual interactions.
• Agent-based models (ABMs) can represent individual-level interactions and demographics but have many unknown parameters.
• Existing data assimilation techniques are often applied to compartmental models, limiting their ability to capture complex social structures.

Purpose of the Study:

• To propose and evaluate a novel method for calibrating an agent-based model (ABM) using daily epidemiological data.
• To integrate the strengths of agent-based modeling with ensemble-based data assimilation for disease dynamics.
• To address the challenge of adapting agent populations to coarse-grained data during calibration.

Main Methods:

• Coupling an epidemiological agent-based model with the ensemble Kalman filter (EnKF).
• Developing two stochastic strategies to correct model predictions and adapt agent populations.
• Utilizing the EnKF with perturbed observations for joint estimation of disease states and epidemiological parameters.
• Applying the methodology to a COVID-19 agent-based model.

Main Results:

• The proposed methodology successfully calibrates an agent-based model using daily epidemiological data.
• Demonstrated the ability to adapt agent populations to incorporate information from coarse-grained data.
• Achieved joint estimation of disease states and key epidemiological parameters through data assimilation.

Conclusions:

• The coupling of agent-based models with ensemble Kalman filter provides a powerful framework for understanding disease dynamics.
• This approach enhances the representation of individual interactions in epidemiological modeling.
• The developed data assimilation techniques offer a robust method for calibrating complex agent-based models with real-world data.