When artificial intelligence turns to drugs

The NGO Drugs for Neglected Diseases Initiative, which seeks treatments for neglected diseases, launched a partnership in April with BenevolentAI, a British company that works to develop new molecules using artificial intelligence (AI). BenevolentAI is not at its first attempt. She notably highlighted, during the pandemic, the role that a molecule, baricitinib, developed by the Eli Lilly laboratory for another disease, can play in the treatment of Covid-19 patients.

Molecules from artificial intelligence

One would think that AI is now all the rage. But in the pharmaceutical sector, the change is more than cosmetic. At the beginning of 2020, Exscientia, a Scottish start-up, developed with the Japanese pharmaceutical laboratory Sumitomo Dainippon a first molecule “built” by AI, entry into clinical trial.

“It’s not futuristic: artificial intelligence is a methodological approach to data processing, which can be used in several stages of the drug industry’s development process”says Dr. Thomas Borel, director of scientific affairs for the federation of pharmaceutical companies (Leem).

A visit to the Paris premises of the French start-up Iktos, founded in 2016, makes us understand the change of era. Here, there are no microscopes or biology devices, no lab technicians in white coats. But computer screens galore, which will cross masses of health data at a speed that no human brain could reach.

“The idea is to exploit existing data to obtain new interesting molecules, more quickly”explains Yann Gaston-Mathé, the leader of the start-up, which he co-founded in 2016.

For this, his team used a global database containing data from 100 million molecules. From this, “we have trained a model that will automatically generate new molecules, and identify those that will be active on biological targets of interest”portrays Yann Gaston-Mathé.

Iktos has even set up a platform for researching molecules using artificial intelligence, which it provides on subscription to pharmaceutical companies.

Aqemia, a start-up from the École nationale supérieure-PSL, created in 2019, is developing a drug discovery platform using statistical physics inspired by quantum.

“We use so-called generative artificial intelligence”underlines the founder, the researcher Maximilien Levesque. “We invent molecules that will stick to a specific biological target responsible for a disease. Artificial intelligence is fed by physics: we just need to know the physical nature of the molecule and the target to calculate their affinity “he describes.

Large laboratories find their interest in it

If start-ups are at the forefront, the major laboratories are increasingly looking into the issue, and paying the price. The American giant Bristol-Myers Squibb signed an agreement with Exscientia last year, to which it could pay more than a billion dollars. The Gafam are also in the game: in 2019, the Swiss laboratory Novartis and the giant Microsoft announced their collaboration on the subject.

Is this the end of the chemist in his laboratory? There are major challenges, including access to actionable data. Without forgetting the need to find future data specialists, experts in both artificial intelligence and pharmacology.

There is also an important regulatory aspect, judge Thomas Borel, of Leem. “For example, we use AI to create a virtual arm of patients during a clinical trial. But for this drug to be accepted, the regulatory systems must recognize the value of the algorithm”he said.

“Drugs have been designed with the help of computers for years,” comments for his part Yann Gaston-Mathé, who says he wants to bring “additional tools for chemists, without wanting to replace man with machine”.