Mixed bubbles: Maximizing epidemics with geographic and categorical assortativities in modular geometric graphs
Conference Year
2023
Abstract
Some of the most influential results in network science show how important a few random long-range links can be on the connectivity of otherwise local networks [1, 2]. In reality, these long-range links are rarely random and instead stem from alternate connection mechanisms. One notable example concerns pathogen-spreading networks dominated by two types of assortativity: geographic proximity (e.g., neighbors or colleagues) and in-community preferences (e.g., travel to family or conferences). We propose a mixture model to explore epidemic dynamics on networks with varying proportions of geometric and categorical assortativity. We combine soft random geometric networks [3] with a community-specific configuration model. Simulations of Susceptible-Infectious-Recovered dynamics (SIR) on our network model show that outbreaks are small when networks are highly assortative in terms of either geographic location or category, and quickly rise in scale when “shortcuts” in the system are provided by mixing mechanisms. Specifically, when the number of communities is large, the worst outbreaks occur when both types of assortativity are equally important. With fewer communities, we find that outbreaks are maximized when categorical assortativity is relatively more dominant than geographic proximity. These results inform many important real-world scenarios. For the ongoing COVID-19 pandemic, our contact networks could maximize spread when they feature an equal mix of assortativity with local communities (neighbors and colleagues) and distributed ones (family). Our work is of particular interest in engineered systems, such as the livestock supply chains, which often mix opportunistic local assortativity with company-driven long range connections. We focus on such engineered systems as our model can suggest structural interventions which could be enacted through laws, incentives, or corporate action. We therefore use our abstract network model to generate hypotheses and interventions which we then test on a large, complex, agent-based model of the swine production network in the United States.
Primary Faculty Mentor Name
Laurent Hébert-Dufresne
Status
Graduate
Student College
College of Engineering and Mathematical Sciences
Program/Major
Computer Science
Primary Research Category
Engineering and Math Science
Mixed bubbles: Maximizing epidemics with geographic and categorical assortativities in modular geometric graphs
Some of the most influential results in network science show how important a few random long-range links can be on the connectivity of otherwise local networks [1, 2]. In reality, these long-range links are rarely random and instead stem from alternate connection mechanisms. One notable example concerns pathogen-spreading networks dominated by two types of assortativity: geographic proximity (e.g., neighbors or colleagues) and in-community preferences (e.g., travel to family or conferences). We propose a mixture model to explore epidemic dynamics on networks with varying proportions of geometric and categorical assortativity. We combine soft random geometric networks [3] with a community-specific configuration model. Simulations of Susceptible-Infectious-Recovered dynamics (SIR) on our network model show that outbreaks are small when networks are highly assortative in terms of either geographic location or category, and quickly rise in scale when “shortcuts” in the system are provided by mixing mechanisms. Specifically, when the number of communities is large, the worst outbreaks occur when both types of assortativity are equally important. With fewer communities, we find that outbreaks are maximized when categorical assortativity is relatively more dominant than geographic proximity. These results inform many important real-world scenarios. For the ongoing COVID-19 pandemic, our contact networks could maximize spread when they feature an equal mix of assortativity with local communities (neighbors and colleagues) and distributed ones (family). Our work is of particular interest in engineered systems, such as the livestock supply chains, which often mix opportunistic local assortativity with company-driven long range connections. We focus on such engineered systems as our model can suggest structural interventions which could be enacted through laws, incentives, or corporate action. We therefore use our abstract network model to generate hypotheses and interventions which we then test on a large, complex, agent-based model of the swine production network in the United States.