Higher-order Interaction Matters: Dynamic Hypergraph Neural Networks for Epidemic Modeling

TL;DR

A novel Human Contact-Tracing Hypergraph Neural Network framework tailored for epidemic modeling called EpiDHGNN is introduced, leveraging the capabilities of hypergraphs to model intricate, higher-order relationships from both location and individual level.

Abstract

The ongoing need for effective epidemic modeling has driven advancements in capturing the complex dynamics of infectious diseases. Traditional models, such as Susceptible-Infected-Recovered, and graph-based approaches often fail to account for higher-order interactions and the nuanced structure pattern inherent in human contact networks. This study introduces a novel Human Contact-Tracing Hypergraph Neural Network framework tailored for epidemic modeling called EpiDHGNN, leveraging the capabilities of hypergraphs to model intricate, higher-order relationships from both location and individual level. Both real-world and synthetic epidemic data are used to train and evaluate the model. Results demonstrate that EpiDHGNN consistently outperforms baseline models across various epidemic modeling tasks, such as source detection and forecast, by effectively capturing the higher-order interactions and preserving the complex structure of human interactions. This work underscores the potential of representing human contact data as hypergraphs and employing hypergraph-based methods to improve epidemic modeling, providing reliable insights for public health decision-making.

Figures (4)

• Figure 1: Illustration of how various graph encoding methods can be employed to capture complex interactions. Hypergraphs, in particular, offer significant advantages over traditional graphs by retaining both individual-level and location-level information, while also capturing higher-order interactions. This enriched representation facilitates a more nuanced understanding of epidemic dynamics.
• Figure 2: Model Architecture of proposed EpiDHGNN model. The arrows in the top left corner refers to the three time stamps defined in Section 3.3, where $[0:tsh]$ is the black interval, $[tsh:ks]$ is the orange interval, and $[ks+1:ps]$ is the green interval. All individual state is masked to 0 in $[0:tsh]$ as shown in the top left black module. Corresponding inputs for source detection and forecast defined in Section 3.3.1 and 3.3.2 is then feed to the model as input. The light blue HyperConv module in defined in Section 4.1; the dark blue temporal convolution module is defined in Section 4.1; and the contact pattern awareness module is defined in Section 4.2.
• Figure 3: Visualization of various $\alpha$'s impact on source detection performance
• Figure 4: Forecast generalizability analysis. The models can successfully the future infection dynamics within various PS. We also provide the Mean Absolute Error (MAE) of Naive Model (A naive time series model forecasts future values by assuming they will be the same as the most recent observed value) and our method.