Uncertainty quantification in mechanistic epidemic models via cross-entropy approximate Bayesian computation.

Area of Science:

• Epidemiology
• Computational Biology
• Statistical Modeling

Background:

• Accurate parameter estimation and uncertainty quantification are crucial for effective epidemic modeling.
• Existing methods may struggle with initial conditions and prior knowledge integration.
• Real-time epidemic forecasting requires robust and data-driven approaches.

Purpose of the Study:

• To propose a novel data-driven approximate Bayesian computation (ABC) framework for parameter estimation and uncertainty quantification in epidemic models.
• To enhance epidemic modeling by incorporating initial condition identification and informative prior learning.
• To demonstrate the framework's efficacy using real-world COVID-19 data.

Main Methods:

• Developed a data-driven approximate Bayesian computation framework.
• Integrated initial condition identification using compatible dynamic states.
• Employed the cross-entropy method for learning informative prior distributions.
• Utilized an ordinary differential equation (ODE) based generalized SEIR model with time-dependent parameters.

Main Results:

• Successfully estimated twelve parameters for a COVID-19 model using Rio de Janeiro data.
• The calibrated model accurately described observed hospitalization and death data.
• The framework demonstrated consistency with observational data.
• Achieved reliable short-term (few weeks) forecast extrapolations.

Conclusions:

• The proposed data-driven ABC framework offers a powerful tool for real-time epidemic modeling.
• The methodology effectively handles parameter estimation and uncertainty quantification.
• The approach is valuable for public health decision-making during epidemics.