With governments in a mad scramble to identify the policies most likely to curb the spread of the pandemic without unnecessarily crippling the global economy, researchers are turning to AI and high-performance computing to analyze how the virus spreads – and how different policies can slow it down. Now, researchers at the University Carlos III de Madrid, Barcelona Supercomputing Center and the Carlos III Institute of Health and CIBER en Epidemiología y Salud Pública have tested a new implementation of an epidemiological model designed specifically to provide a scalable solution for analyzing COVID-19.

The tool, EpiGraph, was introduced in 2011 as a way of modeling the propagation of the flu virus. To do that, the researchers used demographic data and individual interactions extracted from social networks to create a “realistic interconnection network” comprising every single individual in a population. The interconnection network, which only accounted for factors like profession, age and gender at the time, was then fed to a “scalable, fully distributed simulator” alongside a model of the infectious agent. The simulator then examined how the virus traveled within the network of people to determine a probability of infection. 

At the time, EpiGraph compared well to state-of-the-art flu epidemiology tools. Since 2011, though, EpiGraph has considerably evolved – in 2013, for instance, researchers added geographic location and transportation modules to EpiGraph, allowing it to model incoming population and travel between urban areas. Now, EpiGraph has received its highest-profile upgrade yet. 

The new implementation of EpiGraph includes even more detailed parameters: individuals can be characterized as students, workers, stay-at-home or elderly, and the epidemiological elements of the model now include interactions between the disease and climate- and weather-related factors such as temperature and humidity. 

All of this, of course, is intended to tackle epidemiological analysis for COVID-19 – and it’s also designed to scale for high-performance computing. Co-developed by María-Cristina Marinescu, a postdoctoral researcher at the Barcelona Supercomputing Center, EpiGraph is a network I/O-bound application implemented on MPI and intended specifically for large-scale simulations.

To test the new EpiGraph implementation, the researchers studied the spread in Spain – their home, and one of the hardest-hit countries. After validating EpiGraph against actual COVID-19 deaths in Spain, the researchers modeled a number of policy options, including school closures, transportation restrictions (including both short- and long-distance trip restrictions), a higher percent of people working from home and social distancing (wherein the researchers limited leisure connections between all individuals).

The simulations are striking. In a scenario without any restrictions, the number of infections reaches seven million – compared to just over one million when policies are enforced (as Spain has done in reality). 

Next, the researchers tested a scenario where, at week 14, 10% of the population returned to work, but prevention measures were still taken by the entire population. A smaller second peak emerges, with infections rising from around 100,000 to around 300,000.

Their final simulation was somewhat more alarming: when 10% of the workforce was allowed to return to work after 15 weeks of restrictions but prevention measures were not taken, infections – which had fallen to the tens of thousands – spiked back up over a million. 

Now, the researchers plan to analyze a series of short- and medium-term questions, such as gauging the effects of closing borders and identifying the best strategies for reducing the chance of subsequent peaks in infections.