Gaussian Process Emulation for Exploring Complex Infectious Disease Models.

Epidemiological models that aim for a high degree of biological realism by simulating every individual in a population are unavoidably complex, with many free parameters, which makes systematic explorations of their dynamics computationally challenging. This study investigates the potential of Gaussian Process emulation to overcome this obstacle. To simulate disease dynamics, we developed an abstract individual-based model that is loosely inspired by dengue, incorporating some key features shaping dengue epidemics such as social structure, human movement, and seasonality. We trained three Gaussian Process surrogate models on three outcomes: outbreak probability, maximum incidence, and epidemic duration. These surrogate models enable the rapid prediction of outcomes at any point in the eight-dimensional parameter space of the original model. Our analysis revealed that average infectivity and average human mobility are key drivers of these epidemiological metrics, while the seasonal timing of the first infection can influence the course of the epidemic outbreak. We use a dataset comprising more than 1,000 dengue epidemics observed over 12 years in Colombia to calibrate our Gaussian Process model and evaluate its predictive power. The calibrated Gaussian Process model identifies a subset of municipalities with consistently higher average infectivity estimates, which show notable overlap with previously reported dengue disease clusters, suggesting that statistical emulation can facilitate empirical data analysis. Overall, this work underscores the potential of Gaussian Process emulation to enable the use of more complex individual-based models in epidemiology, allowing a higher degree of realism and accuracy that should increase our ability to control diseases of public health concern.