Interpretable and Adaptable Graph Neural Networks: Modeling Metapopulation Dynamics of Vaccine-Preventable Childhood Infections

/Abstract The overarching goal of this proposal is to contribute to the fundamental understanding of the spatiotemporal transmission dynamics and control of vaccine-preventable childhood infections. The central scientific premise motivating this project is that integrative modeling approaches, effectively combining traditional mechanistic models and machine learning techniques, have the greatest potential to address the persisting methodological challenges in fully characterizing and accurately predicting the complex spatiotemporal transmission dynamics of childhood infections. We hypothesize that such integrative approaches can overcome the limitations in the inferential and predictive capabilities of existing models, precisely disentangling and predicting the influence of the underlying drivers of transmission dynamics. As such, they can be leveraged to define the most effective control strategies, including vaccination. Three specific aims are proposed. Aim 1: Develop and validate a highly interpretable and predictive modeling framework by deeply integrating graph neural networks with compartmental models. We deliberately develop a highly flexible and generalized graph neural network-based approach to incorporate the underlying network of infection transmission among interconnected populations (or a metapopulation) such as a collection of neighboring towns. We propose new inferential approaches leveraging Approximate Bayesian Computation to facilitate the integration. Aim 2: Develop a comprehensive software platform to facilitate implementations of our models developed in Aim 1, and to promote and catalyze the development and adoption of machine learning modeling research in the community of infectious disease modeling. Aim 3: Test our developed modeling framework using rotavirus and measles as examples and illustrate how our trained models can be used to guide efficient adaptive vaccination and control strategies for rotavirus and measles in the US. Rotavirus and measles are example types of vaccine-preventable diseases, endemic and re-emerging, respectively. Results of this project will also broadly lend valuable insights into future studies (e.g., designing effective vaccination programs) for other vaccine-preventable diseases (VPDs), such as RSV and norovirus which are expected to become VPDs soon. Project Number: 1R01AI189829-01 | Fiscal Year: 2025 | NIH Institute/Center: National Institute of Allergy and Infectious Diseases (NIAID) | Principal Investigator: Siu Yin (Max) Lau (+1 co-PI) | Institution: EMORY UNIVERSITY, ATLANTA, GA | Award Amount: $711,555 | Activity Code: R01 | Study Section: Analytics and Statistics for Population Research Panel B Study Section[ASPB] View on NIH RePORTER: https://reporter.nih.gov/project-details/1R01AI18982901