Gaussian process emulation for exploring complex infectious disease models.

Area of Science:

• Epidemiology
• Computational Biology
• Statistical Modeling

Background:

• Complex epidemiological models with high biological realism are computationally challenging for parameter exploration.
• Individual-based models (IBMs) offer realism but suffer from complexity and numerous parameters.
• Dengue epidemics are influenced by factors like social structure, human movement, and seasonality.

Purpose of the Study:

• To demonstrate Gaussian Process (GP) emulation as a method to overcome computational challenges in complex epidemiological models.
• To develop and utilize GP surrogate models for rapid prediction of key epidemiological metrics.
• To identify key drivers of disease dynamics and calibrate models with real-world epidemic data.

Main Methods:

• Developed an abstract individual-based model (IBM) inspired by dengue dynamics.
• Focused on three epidemiological metrics: outbreak probability, maximum incidence, and epidemic duration.
• Trained three Gaussian Process (GP) surrogate models to approximate IBM outcomes across an eight-dimensional parameter space.

Main Results:

• GP surrogate models enabled rapid prediction of epidemiological metrics within the IBM's parameter space.
• Identified average infectivity and human mobility as key drivers of epidemic metrics.
• Seasonal timing of initial infection influences epidemic outbreak progression.
• Calibration with over 1,000 dengue epidemics in Colombia validated GP model predictive power.

Conclusions:

• Gaussian Process emulation significantly enhances the utility of complex IBMs in epidemiology.
• This approach allows for greater biological realism and accuracy in disease modeling.
• Statistical emulation facilitates empirical data analysis, aiding in the identification of high-risk areas and improving disease control efforts.