Biology-Informed Recurrent Neural Network for Pandemic Prediction Using Multimodal Data.

Area of Science:

• Biomedical informatics
• Epidemiology
• Artificial intelligence

Background:

• The time from infection to diagnosis often follows a log-normal distribution.
• Accurate pandemic prediction requires modeling social contact and disease transmission dynamics.
• Estimating the true number of infected individuals is vital for effective pandemic control.

Purpose of the Study:

• To propose a novel bio-inspired neural network for pandemic prediction.
• To predict both confirmed and infected cases using multimodal data.
• To evaluate the model's performance against existing epidemiological and AI methods.

Main Methods:

• Developed a back-projection infected-susceptible-infected-based long short-term memory (BPISI-LSTM) neural network.
• Integrated disease data with migration information to model social contact.
• Utilized COVID-19 datasets from India, Austria, and Indonesia for evaluation.

Main Results:

• The BPISI-LSTM model demonstrated superior performance in predicting confirmed cases compared to vSIR and LSTM.
• The model achieved accurate short-term and long-term predictions.
• Mobility data integration enhanced prediction accuracy.
• The model successfully estimated the number of infected cases, an unobservable metric.

Conclusions:

• The proposed bio-inspired BPISI-LSTM model offers a superior approach to pandemic prediction.
• Incorporating mobility data significantly improves forecasting accuracy.
• Estimating infected cases provides critical insights for pandemic management and control.