A Deep Learning-Enhanced Compartmental Model and Its Application in Modeling Omicron in China

Area of Science:

• Epidemiology
• Computational Biology
• Artificial Intelligence

Background:

• Traditional compartmental models for infectious disease transmission require extensive data for parameterization, which is costly and resource-intensive.
• These models often fail to fully capture the complex temporal, spatial, and mobility dynamics of disease spread.
• Existing methods struggle with the data particularity and resource demands for accurate epidemiological forecasting.

Purpose of the Study:

• To explore deep learning techniques as an alternative for estimating stochastic transmission parameters in epidemiological models.
• To develop a model that integrates temporal, spatial, and mobility dimensions for enhanced disease spread prediction.
• To assess the efficacy of deep learning in forecasting the Omicron epidemic in China.

Main Methods:

• Utilized Deep Neural Network (DNN) and Long-Short Term Memory (LSTM) techniques for parameter estimation.
• Developed a customized compartmental model incorporating deep learning-estimated parameters.
• Applied the model to predict Omicron epidemic development in China from June 4 to July 1, 2022.

Main Results:

• Achieved high prediction accuracy: 98% for infections and 92% for deaths.
• Demonstrated deep learning's ability to effectively model temporal, spatial, and mobility aspects of disease transmission.
• Successfully predicted the Omicron epidemic's trajectory over a 28-day period.

Conclusions:

• Deep learning methodologies offer a viable and efficient alternative to traditional compartmental models for infectious disease dynamics.
• The study highlights the potential of DNN and LSTM in epidemiological forecasting, reducing data dependency and improving accuracy.
• This research validates the application of advanced AI techniques for predicting infectious disease spread and informing public health strategies.