Modeling epidemic dynamics using Graph Attention based Spatial Temporal networks.

Area of Science:

• Epidemiology
• Computational Biology
• Data Science

Background:

• Accurate epidemic forecasting is crucial, especially during pandemics like COVID-19 and influenza outbreaks.
• Traditional time-series models struggle to capture complex spatial and temporal disease dynamics.
• Graph neural networks and machine learning show promise but face challenges in data quality and transferability.

Purpose of the Study:

• To introduce and validate the Graph Attention-based Spatial Temporal (GAST) model for enhanced epidemic forecasting.
• To address limitations of existing models, including data quality and overfitting issues.
• To provide a robust framework for accurate short-term, daily infectious disease spread predictions.

Main Methods:

• Development of the Graph Attention-based Spatial Temporal (GAST) model utilizing graph attention networks (GATs).
• Integration of spatio-temporal analysis for a nuanced understanding of epidemic dynamics.
• Validation using COVID-19 and influenza datasets for forecasting performance.

Main Results:

• The GAST model demonstrates superior forecasting capabilities compared to traditional methods.
• The model provides accurate short-term, daily predictions for both COVID-19 and influenza.
• Validation highlights the model's effectiveness across diverse geographical data.

Conclusions:

• The GAST model offers a more adaptable and robust approach to epidemic forecasting.
• Its sophisticated spatio-temporal analysis framework enhances understanding of disease spread.
• The GAST model has the potential to significantly inform public health interventions for various infectious diseases.