Using our data and know-how to aid decision-making during a pandemic


From research into possible treatments to mobile apps and anticipating future epidemics, data science is at the heart of the scientific, economic and social challenges raised by the pandemic. At Université PSL, laboratories and research teams are mobilizing to mount a response to this urgent health crisis. Jamal Atif, a professor at Dauphine–PSL and Deputy Scientific Director at the Paris Artificial Intelligence Research Institute (Prairie), and his team are currently working on developing a decision-making tool for use in managing the COVID-19 epidemic. Here’s our conversation.

PSL: Could you tell us about some of the projects that PSL’s data science research teams are tackling right now to address the COVID-19 crisis?

Jamal Atif: During this unprecedented global crisis, it’s impossible to overstate the critical role being played by digital technology. From remote working to online shopping and mobile apps, we’re seeing the emergence of a post-crisis world where digital tech will be even more ubiquitous. And digital research is no exception. Epidemiological models are nothing new, but this is the first time we’ve worked on massive volumes of data to address a crisis, at least in Europe. And data science methods and tools are what’s made that possible. PSL is conducting a host of research projects in that area: the use of machine learning to discover promising chemicals, imaging-based diagnostics, predictive models based on mobility data and so on. It would be impossible to cite every project, especially since I only have partial knowledge of each one. For my part, alongside other PSL researchers (from Dauphine–PSL, MINES ParisTech–PSL and ENS–PSL) and partners including the CNRS, I’m hard at work on research into decision-making tools that use mobility tools for managing the COVID-19 epidemic. The Prairie Institute has been a big help in that regard, in part by giving us access to data from one of its partners but particularly by raising interest among colleagues and getting them involved in the project. What we hope to do as quickly as possible is provide tools that governments and the general public can use to get a better understanding of the pandemic and identify the health policies that will be most effective.

PSL: How exactly are these research projects aiding the government’s efforts to manage the current health crisis?

Jamal Atif: There’s a real spirit of community above all. Everyone is willing to pitch in. Even high-level colleagues are taking part. One example is Emmanuel Bacry, who’s the CNRS Senior Research Fellow at CEREMADE (Dauphine–PSL) and the Chief Scientific Officer for the Health Data Hub (HDH). The French government has asked him, along with the director of the HDH, to lead a taskforce on how data, and health data in particular, can be used to try to curb the pandemic.

We try to quantify every model’s level of uncertainty. We always come back to the famous saying: All models are wrong, but some are useful.

In our own project, after a tedious process of compiling and preparing data – which we haven’t quite completed – we’re in the process of finalizing our models. The next steps will be to discuss our models and results with epidemiologists, and if they prove to be relevant, we’ll forward them to CARE, the Committee for Analysis, Research and Expertise appointed by French President Emmanuel Macron on March 24 and chaired by the Nobel Laureate in medicine Françoise Barré-Sinoussi, and to its equivalent at the CNRS (care@CNRS).

But bear in mind that we need to be cautious. As scientists, we don’t take prescriptive action. We work on models that have known limitations; we simply try to evaluate their potential. It would be boastful, imprudent and contrary to scientific method to claim any certainty. I have to stress that point. In science, your knowledge is never set in stone, and in this case, the deeper we dig, the more gray areas we find. Our datasets are heterogeneous and sometimes incomplete, and they’re often tainted with bias and noise; the main challenge we face is in creating datasets we can use, with help from organizations and other partners, and that’s not an easy task. And we try to quantify every model’s level of uncertainty. We always come back to the famous saying: All models are wrong, but some are useful.

PSL: Major tech firms like Facebook, Google and Orange are making their data and infrastructure available to scientists and engineers around the world so they can develop tools for leading us out of the crisis. Do you see that as an opportunity for PSL’s scientists?

Those are legitimate concerns, and we’ve been very strict in that regard. We use spatially aggregated data that doesn’t compromise user privacy.

Jamal Atif: It’s important information that’s being provided to everyone in the world of research and that could be combined with other datasets. At PSL, for example, we very quickly gained access to the data in Facebook’s Data For Good program. For the project I’m working on, that data will help us study the impact of mobility on predicting future pandemics. These days, as soon as you mention the big tech firms and data in the same sentence, everyone worries about their personal privacy. Those are legitimate concerns, and we’ve been very strict in that regard. We use spatially aggregated data that doesn’t compromise user privacy. In other words, we’re not interested in knowing where a particular person is going; we use pre-quarantine mobility data and up-to-the-minute health data to make localized projections of the pandemic’s possible spread in France, based on various scenarios for lifting the lockdown. We also hope to use that information to determine whether covariables for mobility and density have real predictive power about the epidemic’s spread, and possibly even advise the general public once the lockdown has been lifted.