Using big data to precisely and easily predict an epidemic’s course or a virus’s spread—that’s the seemingly incredible goal of Jean-Charles Delvenne, a researcher at the Mathematical Engineering Centre of UCL’s Institute of Information and Communication Technologies, Electronics and Applied Mathematics. At stake: better management of major epidemics.
When an epidemic such as Zika strikes South America, or a nosocomial disease spreads in a hospital, predicting its future is as essential as providing immediate care. Indeed, knowing how and at what speed a virus or bacteria will spread helps determine the most efficient way to contain an outbreak: a line of attack that involves not only doctors but experts in the analysis of information communications.
Extremely rich databases
In recent years such experts have had to deal with a new parameter: databases that supply information as plentiful as it is precise. This is big data. ‘The means of data capture have evolved as much as the means of data storage’, Prof. Delvenne explains. ‘The result is that we have very detailed information on populations: their locations, how they communicate, their movements, etc. So, compared to the past, we’re less and less restricted by the availability of information. That’s good news! However, it changes how we work: we don’t analyse such databases as quickly and effectively as we do smaller databases.’ Making these new databases genuinely useful requires new techniques for analysing them.
Social network vs. social dynamics
Assisted by his colleagues Luis Rocha and Renaud Lambiotte of the University of Namur, Prof. Delvenne decided to tackle the problem through the prism of two types of data:
- the social network of individuals;
- the social dynamics of individuals.
Two essential elements never considered simultaneously. ‘The social network is how individuals are interlinked. Its structure can be a school, a city, a country, etc. This information has been studied for ten years in academic circles and it’s enabled us to learn which segments of the population will be most affected by an outbreak. The social dynamics of individuals is tied to the frequency with which they meet, talk, touch as well as the regularity of these contacts. These social dynamics directly influence the order and timing of the transmission of a virus, for example.’
Less is more
Once the stage is set, you still have to know how to effectively use both types of data, especially since the more information you have, the longer its analysis takes. ‘When we analyse an epidemic we have to be able to test many scenarios within a reasonable time frame’, insists Prof. Delvenne. ‘We have to be effective as soon as possible.’ To this end, one of his team’s observations is crucial. ‘Our work allowed us to observe that even though these two parameters are important, they’re not necessarily useful at the same time. Very often, one outweighs the other. So sometimes it’s the social network that provides the more valuable information, other times it’s the social dynamics.’
Determining who in the network or in the dynamics takes precedence
With this observation in mind, rather than developing a tedious algorithm, which would take a lot of time to analyse the two parameters simultaneously within a dataset, Prof. Delvenne and his colleagues chose a quicker, more effective option: develop an algorithm that makes it possible to quickly predict who in the network or in the dynamics takes precedence. ‘Once this prediction is made,’ Prof. Delvenne concluded, ‘it’s enough to launch a second algorithm relatively quickly that will predict, as precisely as possible based on the most relevant parameter, how the epidemic will develop.’
Elise Dubuisson
A Glance at Jean-Charles Delvenne's bio
