DSIH, TUESDAY OCTOBER 25, 2022
MIT, the National Academy of Medicine and the Health Data Hub (HDH) organized the second edition of the symposium on October 20 “Artificial Intelligence and Medicine: promises and limits”during which American and European health AI experts discussed concrete use cases.
This half-day was structured around three sessions aimed at shedding light on the AI approaches applied to health on both sides of the Atlantic, but also at discussing the possibilities and practical realities of the latter.
The development of new drugs is one of the sectors where AI is already playing a role. This is indeed a huge challenge, requiring a commitment of 10 to 15 years and an investment of up to $2 billion. Artificial intelligence algorithms can reduce these expenditures of time and money, testified Nicolas Do Huu, co-founder and Chief AI Officer at Iktos, a French start-up specializing in the development of AI solutions applied to chemical research, more particularly to medicinal chemistry, and to the design of new drugs.
Nicolas Do Huu described the different applications of these solutions: predictive modeling of the activity of the molecule studied, generative modeling, which creates data, learning by reinforcement, so that each new iteration is more efficient than the previous, virtual screening and 3D/4D simulation. “In five to eight years, a drug will have been developed using AI”, he predicted. In his eyes, this is a revolution comparable to that experienced by the automotive industry with automation.. “In the near future, we will have the same kind of drug factory that we have for Teslas today”, he added. An enthusiasm shared by Farhad Rikhtegar Nezami, a researcher at Harvard Medical School, who, during his speech, confirmed in the form of a joke: “No one believes in molecular simulation except those who do it. »
Leo Anthony Celi, researcher at MIT and Harvard Medical School, then presented his work on the search for new therapeutic indications for existing drugs through the analysis of mass data, including computerized patient records. He nevertheless warned about the biases in the data for which RNs should be trained: missing information, gender of the patient, etc. “There are many opportunities for new therapeutic indications, but we must be aware of these biases”he said, also citing the ” expiration date “ algorithms that may become obsolete.
The contribution of AI to clinical trials was described by Stéphanie Allassonnière, professor and vice-president of Valorisation at the University of Paris Cité and associate professor at the École polytechnique. “We can create digital twins”, but also, with enough data, “trajectories allowing us to classify a patient, to see how he reacts to a molecule and to anticipate his reactions over several years”, she clarified. Artificial intelligence also makes it possible to create “false patients, to increase the cohorts where it is difficult to recruit and thus improve the clinical trial”.
Jean-Yves Blay, head of the medical oncology department of the Léon-Bérard center and president of Unicancer, also cited the possibility offered by AI to predict the potential toxicity of a trial for patients, thus making it possible to put an end to it as soon as possible. The president of Unicancer nevertheless ended his speech by recalling the need for the authorities to be able to adapt in order to assess these new tools, the development of which is very rapid, at the risk, otherwise, of accumulating late.
We would like to give thanks to the author of this post for this remarkable material
Artificial intelligence: already concrete applications in health
You can view our social media profiles here and other related pages herehttps://www.ai-magazine.com/related-pages/