Neural parameter calibration for dengue outbreak forecasting.

Area of Science:

• Epidemiology
• Computational Biology
• Mathematical Modeling

Background:

• Dengue fever is a significant public health issue in tropical and subtropical areas.
• Complex interactions between viral and host factors drive dengue transmission dynamics.
• Computational models, often using ordinary differential equations (ODEs), are crucial for understanding these dynamics.

Purpose of the Study:

• To evaluate Neural Parameter Calibration (NPC) for estimating parameters in an Extended Compartment Model (ECM) of dengue transmission.
• To compare the computational efficiency and accuracy of NPC against traditional Markov Chain Monte Carlo (MCMC) methods.
• To validate the ECM-NPC approach using real-world dengue surveillance data from South America and Southeast Asia.

Main Methods:

• Developed an Extended Compartment Model (ECM) comprising seven ODEs to describe human and mosquito dengue transmission.
• Employed Neural Parameter Calibration (NPC), utilizing neural networks to learn model parameter posterior distributions.
• Analyzed six dengue surveillance datasets from three South American cities and three Southeast Asian countries.

Main Results:

• NPC demonstrated significantly faster computation times compared to MCMC (e.g., 368s vs. 2998s for country-level data).
• NPC achieved comparable accuracy to MCMC, with lower Mean Squared Error (MSE) values for both city and country datasets.
• The combined ECM and NPC approach proved effective for dengue outbreak forecasting.

Conclusions:

• The integration of Extended Compartment Models with Neural Parameter Calibration provides accurate dengue outbreak forecasts.
• This approach offers a substantial reduction in computational cost, making it a practical tool for public health.
• The ECM-NPC method is particularly valuable for supporting timely interventions in resource-limited settings.