Amazon Web Services Inc. today introduced an open-source toolkit that researchers can use to simulate how COVID-19 spreads in communities and estimate what results different kinds of government responses might produce.

The toolkit includes a simulator that can make projections about how many COVID-19 cases will be reported in a region over a given time frame. There are also machine learning models that help with error reduction and other tasks involved in producing projections.

According to AWS, the toolkit enables researchers to run what-if experiments simulating COVID-19 outbreaks in different regions. It’s possible, for example, to simulate a scenario where a second wave takes place six months after the first with a different number of daily cases. Researchers can also estimate how a mild government response would compare with a stricter response in terms of its effectiveness.

“The output is the projection of the total confirmed cases over a specific timeline for a target state or a country, for a given degree of intervention,” the team of AWS engineers and executives involved in the project explained in a blog post today.

The toolkit generates case projections by applying statistical methods to historical data provided by the user. Researchers can, for example, base their simulations on recent COVID-19 case counts shared by government agencies in the country or state they’re studying. Simulations can also make use of inferred parameters, such an estimated rate of transmission extrapolated by a machine learning model from hospital statistics. 

“Our solution first tries to understand the approximate time to peak and expected case rates of the daily COVID-19 cases for the target entity (state/country) by analysis of the disease incidence patterns,” the AWS team detailed. “Next, it selects the best (optimal) parameters using optimization techniques on a simulation model. Finally, it generates the projections of daily and cumulative confirmed cases, starting from the beginning of the outbreak until a specified length of time in the future.”

The toolkit is available on GitHub.

Photo: Tony Webster/Flickr