Predicting epidemic evolution on contact networks from partial observations.

Area of Science:

• Computational Epidemiology
• Network Science
• Mathematical Biology

Background:

• Computational models are crucial for network epidemiology and epidemic forecasting.
• Belief Propagation (BP) is an established efficient method for identifying epidemic origins in compartment models like Susceptible-Infected-Recovered (SIR).

Purpose of the Study:

• To investigate the efficacy of Belief Propagation for predicting the future evolution of epidemic outbreaks.
• To compare BP's predictive performance against Monte Carlo direct sampling using early-stage partial observations.

Main Methods:

• Application of Belief Propagation to the SIR model on various graph structures (random, power-law, real-world networks).
• Comparison of BP predictions with Monte Carlo direct sampling under different observation strategies.
• Analysis of prediction quality based on observed data and target metrics (e.g., epidemic size, extinction time).

Main Results:

• Belief Propagation generally provides superior predictions compared to direct sampling for epidemic forecasting.
• Prediction accuracy is influenced by the specific metric being evaluated (e.g., individual states, epidemic size).
• The quantity of observed infected nodes at the early stage significantly impacts prediction quality.

Conclusions:

• Belief Propagation is a powerful and reliable method for epidemic forecasting in network models.
• The study validates BP's utility beyond origin identification, extending it to future state prediction.
• Optimizing early-stage data observation is key to maximizing the accuracy of epidemic forecasts.