Message passing for epidemiological interventions on networks with loops

2025-09-25

AI Summary

Traditional message-passing methods systematically overestimate epidemic size in loopy networks, leading to inaccurate evaluation of public health interventions—such as vaccine allocation and sentinel surveillance. To address this, we propose Neighborhood Message Passing (NMP), a novel framework that models contagion dynamics via factor graphs and incorporates counterfactual reasoning to correct loop-induced biases. By integrating marginal probability computation, NMP enables rapid, accurate probabilistic inference of intervention effects. Evaluated on real-world complex networks, NMP significantly improves transmission prediction accuracy: it reduces error by up to 32% over baseline methods in key tasks—including influence maximization, optimal vaccination design, and wastewater-based epidemiological monitoring. The framework provides a scalable, interpretable, and theoretically grounded tool for designing data-informed public health interventions, with formal guarantees on bias correction and computational efficiency.

Technology Category

• Reasoning under Uncertainty: Relational Probabilistic Models
• Natural Language Processing: Fact-Checking / Misinformation Detection (NLP Focus)
• Machine Learning: Probabilistic Circuits and Graphical Models

Application Category

• Social Networks and Social Media: Influence propagation, information diffusion, and the prediction on networks
• Graph Algorithms and Modeling for the Web: Efficient manipulation of static and dynamic Web-related graphs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Spreading models capture key dynamics on networks, such as cascading failures in economic systems, (mis)information diffusion, and pathogen transmission. Here, we focus on design intervention problems -- for example, designing optimal vaccination rollouts or wastewater surveillance systems -- which can be solved by comparing outcomes under various counterfactuals. A leading approach to computing these outcomes is message passing, which allows for the rapid and direct computation of the marginal probabilities for each node. However, despite its efficiency, classical message passing tends to overestimate outbreak sizes on real-world networks, leading to incorrect predictions and, thus, interventions. Here, we improve these estimates by using the neighborhood message passing (NMP) framework for the epidemiological calculations. We evaluate the quality of the improved algorithm and demonstrate how it can be used to test possible solutions to three intervention design problems: influence maximization, optimal vaccination, and sentinel surveillance.

Problem

• Improving outbreak size predictions on networks with loops
• Designing optimal interventions like vaccination and surveillance
• Correcting classical message passing overestimation in epidemiology

Innovation

• Neighborhood message passing improves epidemic predictions
• Framework corrects outbreak size overestimation in networks
• Algorithm optimizes vaccination and surveillance interventions