Lessons from Circletown

An invader has quietly slipped into Circletown and Squaretown.

As the 400 residents of each small town travel to and from work, school, social events, and the hospital, one of them has unwittingly become infected with the deadly scourge of smallpox. Soon, the once all-but-eliminated virus will be spreading through both towns, leading to hospitalizations and deaths.

Fortunately, all the residents of Circletown and Squaretown are virtual. The simulation is part of a new effort by School researchers to use complex computational models to discern patterns in the chaotic dance of infection through a populace.

Spurred in part by last fall’s anthrax attacks, Donald Burke, MD, director of the School’s Center for Immunization Research, struck up a collaboration with Josh Epstein, a researcher at the Brookings Institute, to develop better ways to respond to bioterror attacks through agent-based computational modeling.

“Computational modeling is not focused so much on a statistical output where you get actual numbers, but is instead directed toward detailing patterns of flow,” Burke explains. “You can use it to build simulations where you create a large number of entities that can represent people, microbes, or almost anything else. You then put these entities on a 2-D framework, and allow them to move about in that framework so that they interact with each other.”

According to Burke, the approach may be particularly useful because it allows researchers to simulate highly inhomogeneous populations whose variety in behavior, vulnerability to infection, and other factors are similar to those of real-world populations.

Derek Cummings and Ramesh Singa, two graduate students active in the Burke group’s modeling work, can run the Circletown and Squaretown model on a laptop they keep in a room down the hall from Burke’s office.

Cummings, an MHS student and a doctoral candidate in Geography and Environmental Engineering at Homewood, notes that for each one of the towns’ 800 residents, the researchers can adjust such factors as age, activity levels, social contacts, and susceptibility. Residents who become infected progress through the established stages of the disease, and that has an impact on their chances of passing the infection to others. 

“The reason this is useful is that many times the outcomes of these complex adaptive systems are not at all obvious,” says Burke.

Every time the program is run, chance factors have countless small impacts on the model’s results. But Cummings and Singa, an MHS student, have also been adjusting the overall model to match its results to patterns of infection spread observed in smallpox outbreaks in Europe in the 1970s. In addition, they’ve begun to try quelling the epidemic, providing a means of testing the effectiveness of various public health policies.

Burke notes that similar agent-based techniques are being used in military training, economics re-search, and management of traffic flow in urban areas.

“It’s potentially a highly complementary approach to statistical modeling, which should help supply us with the parameters we use to govern the behavior of individual agents,” Burke says. “No one can promise that this approach will be a substantial advance over existing techniques, but my intuition is that the time to test it is now, given the advances in computational power, epidemiology, and microbiology.”

Burke anticipates submission of a first paper from his group on the new approach’s results early this fall.