DeepDynaForecast: Phylogenetic-informed graph deep learning for epidemic transmission dynamic prediction.

Area of Science:

• Epidemiology
• Computational Biology
• Public Health

Background:

• Accurate prediction of epidemic transmission is crucial for effective public health interventions.
• Phylogenetic trees and phylodynamics are valuable tools for understanding disease spread and identifying high-risk populations.
• Existing methods may lack the precision needed for early identification and prediction of transmission dynamics in emerging risk groups.

Purpose of the Study:

• To demonstrate the utility of phylodynamic trees for transmission modeling and forecasting.
• To develop and validate a novel phylogeny-based deep learning system, DeepDynaForecast, for predicting epidemic transmission dynamics.
• To enable early identification and prediction of transmission in emerging high-risk groups.

Main Methods:

• Development of DeepDynaForecast, a deep learning system utilizing phylodynamic trees.
• Leveraging a primal-dual graph learning structure with shortcut multi-layer aggregation.
• Validation using simulated outbreak data and empirical human immunodeficiency virus (HIV) epidemic data from Florida (2012-2020).

Main Results:

• DeepDynaForecast demonstrated high accuracy in predicting transmission dynamics using simulated data.
• The model effectively utilized empirical HIV epidemic data to showcase its practical utility.
• The system is adept at the early identification and prediction of transmission dynamics in emerging high-risk groups.

Conclusions:

• Phylodynamic trees, when integrated into deep learning frameworks like DeepDynaForecast, offer a powerful approach for epidemic transmission modeling and forecasting.
• DeepDynaForecast provides a robust tool for public health programs to optimize interventions by predicting disease spread in specific risk groups.
• The open-source availability of the framework facilitates broader adoption and further research in epidemic forecasting.