Neural parameter calibration and uncertainty quantification for epidemic forecasting.

Area of Science:

• Epidemiology
• Computational Biology
• Machine Learning

Background:

• Accurate forecasting of contagion dynamics is crucial for pandemic response.
• Policy-making requires uncertainty quantification for effective resource allocation.
• Traditional methods like Markov-Chain Monte Carlo (MCMC) can be computationally intensive.

Purpose of the Study:

• To develop and apply a novel computational method for learning probability densities on contagion parameters.
• To provide uncertainty quantification for pandemic projections using a neural network approach.
• To compare the performance of the novel method against MCMC-based schemes.

Main Methods:

• Utilized a neural network to calibrate an Ordinary Differential Equation (ODE) model.
• Applied the method to COVID-19 spread data from Berlin in 2020.
• Demonstrated convergence on a simplified SIR model and learning capabilities on reduced datasets.

Main Results:

• Achieved significantly more accurate calibration and prediction compared to MCMC.
• Provided meaningful confidence intervals for infection figures and hospitalization rates.
• Neural network training and execution took minutes, compared to hours for MCMC.

Conclusions:

• The novel neural network method offers a faster and more accurate approach to pandemic forecasting and parameter learning.
• Effective uncertainty quantification is essential for informed public health policy.
• The method shows promise for learning complex epidemiological models from limited data.