Amedeo Napoli : " Illness is not the same as a machine malfunction!"
Date:
Changed on 30/03/2020
Artificial intelligences are often spoken of like that, in the plural, as if they were some kind of“wise creatures”. But those who don't know what the term actually covers need to be more careful in how they use it. Computer scientist Amedeo Napoli works in artificial intelligence, data mining and machine learning. He specialises in“knowledge discovery in databases”. Director of Research at CNRS (French National Centre for Scientific Research) and head of the Orpailleur research team–affiliated with both Inria Nancy Grand Est and LORIA–his work has applications in biology, chemistry and medicine, among other fields.
"Back in the days of ‘expert systems’ in the 1980s, it was believed that medical diagnosis was only a matter of time. But it never actually happened, because although fault-diagnosis systems work quite well, human beings are not machines and diagnosing problems in us is significantly more complex !
Whats is the Orpailleur research team working on?
We’re doing data mining. For example, we take a two-variable data table that aligns individuals and characteristics. We look to identify those characteristics that distinguish individuals who suffer, or not, from an illness being studied. Then, once we’ve identified them, we try to find out whether certain distinguishing characteristics or markers are predictive of the illness. Both tasks can be very difficult to accomplish.
What are the objectives of your work?
One of the aims is to undertake more comprehensive analyses in order to be able to set up large-scale screening for conditions. The problem is that we don’t know for how long a marker is going to be discriminatory and/or predictive. The environments in which the world is evolving are in constant flux and, at any given moment, a particular factor can take on pathological significance by combining with certain other markers that have no effects in isolation. Working on marker combinations and how they interact with the environment is hugely complex.
What are you working on at the moment?
Lots of things... Together with nutritionists at the Clermont-Ferrand INRA research centre, we are looking for predictive markers of type 2 diabetes in a cohort containing data on people going back more than 20 years. It's a global study, encompassing factors linked to biology, genetics, people’s lifestyle habits, where they live, etc.
We are also doing research on antibiotics whose efficacy is known to be affected by bacterial resistance, so much so that the World Health Organisation has made it a priority. We’re using databases of existing antibiotics, as well as those of molecules that may potentially become antibiotics. Other molecules are added to these, so that we can see whether they have the same characteristics as antibiotics. So, it's a matter of trying out innovative combinations that may result in new antibiotics, which would then need to be bench-tested. Finding new ways to combine these molecular structures computationally should help advance the state of the art.
Thanks to a system that draws on the principles of artificial intelligence?
These studies stem from knowledge discovery, so they are, in fact, linked to AI. For a long time, data was “at odds” with knowledge. But clearly, there is natural give-and-take between them and that will continue to be the case. On the one hand, data can be analysed in order to reveal elements of knowledge. On the other, knowledge can be introduced into data mining and managed and analysed in the same way as data. These days, it’s best to bridge certain gaps and work together in order to address medical problems whose resolution is difficult but important for society as a whole.”
More
© Re.Med. / Nov 2018
Laurence Verger
Research communication manager
Nancy CHRU (Regional University Hospital)