The omicron variant of Covid-19 is sending cases skyrocketing around the world. Still, many national and local governments are hesitant to disrupt society in major ways as they did in 2020, opting instead to leave schools and businesses open and mitigate the pandemic through more targeted policies. Luckily, those institutions can be better-prepared thanks to computational tools developed over the last two years. One of those tools, called GeoACT, was made possible by supercomputers and which helps San Diego County plan for Covid-safe school operations.

GeoACT originated at the start of the pandemic, with San Diego’s Health and Human Services Agency (HHSA) organizing several Covid projects with researchers at the University of California San Diego. Essentially, GeoACT—short for “geographically assisted agent-based model for Covid-19 transmission”—enables the creation of custom agent-based models for schools in San Diego County.

An agent-based model (ABM) uses a virtual environment with virtual inhabitants (agents) with various characteristics that interact in various ways throughout a simulated period of time. ABMs are particularly useful for viral transmission applications, as they help to model spread of a virus throughout a given population.

The GeoACT team developed a pipeline to digest school floor plans, convert them into GIS files and label them with relevant information (e.g. doorways, outdoor areas, classrooms). This information is then supplemented with a series of inputs from school officials: agent behavior (e.g. whether or not students can congregate in groups or have lunch in the cafeteria); mask-wearing practices (and what kinds of masks, if any); ventilation type; class sizes; seating layouts; and more. The model can even tweak the percent of time that students are whispering, talking loudly, or breathing heavily, and it was built to incorporate changes in testing frequency and vaccination.

With all these inputs in place, the agent-based model can then yield an expected rate of Covid infection after a given period of time, complete with attribution of cases to given variables. “The simulations allowed us to pinpoint areas in schools that would present higher COVID-19 transmission risks, and to evaluate relative importance of non-pharmaceutical interventions such as wearing masks, reducing class sizes or canceling lunch in the cafeteria and moving it to classrooms,” said Ilya Zaslavsky, lead researcher for the project and director of the Spatial Information Systems Laboratory at the San Diego Supercomputer Center (SDSC), in an interview with SDSC’s Kimberly Mann Bruch.

The model was built to operate on the Comet and Expanse supercomputers at SDSC. Comet leverages Intel Haswell CPUs and Nvidia GPUs to deliver 2.76 peak petaflops, while Expanse’s AMD Epyc CPUs and Nvidia GPUs deliver 5.16 peak petaflops. The supercomputers were accessed with the help of the NSF’s Extreme Science and Engineering Discovery Environment (XSEDE). Using an Apache Airavata gateway, the researchers were able to quickly iterate on models and scenarios. Within days of an early demo of the model to educators in late 2020, they ran it hundreds of times in anticipation of school reopenings in early 2021.

“At a time when schools were struggling with how to reopen safely, the SDSC team developed an easy-to-use, free tool that allowed schools to test their plans on pixels rather than pupils,” said Leslie Ray, senior epidemiologist for the HHSA.

To learn more, read the reporting from SDSC’s Kimberly Mann Bruch here and read the paper, which is linked here.