Multi-region infectious disease prediction modeling based on spatio-temporal graph neural network and the dynamic...

Area of Science:

• Epidemiology
• Computational Biology
• Network Science

Background:

• Human mobility significantly influences infectious disease outbreaks across regions.
• Predicting multi-regional epidemic spread using deep learning and mobility data is a key research area.
• Existing models may overlook spatial dependencies in dynamic network structures caused by human movement.

Purpose of the Study:

• To develop a hybrid framework, Metapopulation Graph Transformer Neural Network (M-Graphormer), for high-dimensional parameter estimation and multi-regional epidemic prediction.
• To address limitations in existing models regarding hidden spatial dependencies in dynamic networks.
• To enable accurate prediction of infectious disease transmission dynamics and provide early warnings.

Main Methods:

• Incorporating Graph Transformer Neural Network and graph learning into a metapopulation SIR model.
• Developing the M-Graphormer hybrid framework for dynamic graph structures.
• Performing multi-wave infectious disease prediction using real-world epidemic data.

Main Results:

• M-Graphormer accurately estimates high-dimensional parameters and predicts epidemic transmission dynamics across multiple regions.
• The framework demonstrates effectiveness even with low-quality data.
• Retrospective analysis revealed temporal contact rate patterns under various interventions and enabled early prediction of disease arrival times.

Conclusions:

• M-Graphormer offers a robust solution for multi-regional epidemic prediction and parameter estimation.
• The framework's ability to capture spatial dependencies enhances prediction accuracy.
• Findings support flexible, region-specific intervention strategies and early warning systems for infectious diseases.