Complete British News World

Bild som visar två Europakartor över ett simulerat utbrott av en pandemi.

Machine learning could help slow future epidemics

In this study, researchers developed a way to improve testing strategies in epidemic outbreaks, and with relatively limited information, were able to predict which individuals would be most useful for testing.

This could be a first step toward a better community in its ability to control future major disease outbreaks and reduce the need for major lockdowns, says Laura Natalie, PhD student in physics at the University of Gothenburg and lead author of the published study.

Rapid control of an outbreak

Machine learning is a type of artificial intelligence that can be described as a mathematical model in which computers are “trained” to learn to see connections and solve tasks with the help of various data. The researchers used machine learning to simulate an outbreak of a pandemic, in which information about the first confirmed cases is used to make calculations of infection in the rest of the population. Among other things, data is used about the communication networks of infected individuals: who was in close contact with them, where and for how long.

In the study, it is possible to quickly control an outbreak when using the method, while a random test leads to uncontrolled disease outbreaks with many infected people. Under real conditions, information can be added, such as demographic data, age and health conditions, which can further improve the effectiveness of the method. Additionally, the same method can be used to prevent reinfection in the population, if post-disease immunity is only temporary, says Laura Natalie.

The map images show two possible results

Both maps show simulating a volcanic eruption at the same time, both by researchers’ method and without. The left map indicates a situation where the method was not used, and where no community closure occurred. Orange (contagious) and black (previously infected and now immune or deceased) predominate. The map on the right indicates the position at the same time, but where the researchers’ method was used. Gray (not infected but sensitive) dominates the map and blue, which symbolizes closed areas that could have been rotated thanks to this method. There are no longer any infected individuals who could continue the epidemic.

Photo: Laura Natalie

Laura Natalie asserts that the study is a simulation and that testing with real data is necessary for the further development of the method. At the same time, it sees the research as a first step toward being able to make more specific efforts to limit the spread of infection, as the testing strategy based on machine learning automatically adapts to the specific characteristics of the disease. For example, it indicated that it can be more easily predicted whether a particular age group should be tested, or whether a limited geographical area is in a danger zone – such as a school, urban area, or a particular neighborhood.

More accurate location of the infection

When an outbreak begins in a major way, it is important to quickly and effectively find infected individuals. In randomized testing, there is a high risk that this will not happen, but with a more targeted testing strategy it is possible to find more infected individuals and thus gain the knowledge necessary to limit the spread of infection. We’re showing that machine learning can be used to develop this testing strategy, she says.

See also  Norrbotten District is now intensifying the search for Delta virus in the district

There are a few previous studies looking at how machine learning has been used in epidemics, especially with this clear focus on finding the best testing strategies.

We show that it is possible to start from relatively simple and limited information to make predictions about the most useful individuals for the test. It provides an opportunity to make better use of available testing resources.

Scientific material:

Improve epidemic testing and containment strategies using machine learning.


Laura Natali, PhD student in the Department of Physics, University of Gothenburg, [email protected]