Posted at 12:00 p.m.
The very young pharmaceutical company Valence, established in Montreal and hosted by the Mila research institute, uses artificial intelligence to accelerate the discovery of new drugs. “We are in the process of developing one of the most powerful chemistry engines in the world, to be able to design and optimize molecules,” explains Therence Bois, the company’s director of operations.
Currently, to develop a drug, researchers begin by determining a “therapeutic target”, that is, a protein or substance that has a proven link with a disease. Then, they imagine and synthesize the molecules that could act on this therapeutic target. The problem is that it is very difficult to know in advance which molecule will actually react in the right way: it is therefore necessary to synthesize and test dozens of candidates before falling by chance on the right one; a tedious and very expensive process.
A promising sort
Where artificial intelligence is a game-changer is that you can use it to predict which molecules will do well before you start testing them in the lab.
Through all the trials that are conducted by various companies, there is data that is generated. What we are trying to do in Valence is to use this data to create models that can predict the reactions of molecules.
Therence Bois, director of operations at Valence
Models such as those developed by Valence are not precise enough to determine with certainty which molecules will be the right ones, but they can direct researchers towards more promising molecules and thus greatly reduce the number of laboratory tests necessary before finding the good molecule. They can also help predict whether molecules will be toxic or not – a very useful quality when we know that certain drugs must sometimes be rejected after millions of dollars invested because of undetected side effects.
Work with the minimum
If the artificial intelligence developed by Valence is so innovative, it is because it is particularly difficult to apply AI to drug development. “In other areas, we can base our artificial intelligence on millions of examples. In pharmacology, as clinical trials are very expensive to conduct, there is much less data and therefore traditional deep learning technologies do not work,” explains Therence Bois.
The sinews of war is therefore to adapt your AI so that it works with as little data as possible. To meet this challenge, Valence can count on the expertise of the experts of the Mila Institute, within which it has its household. “Our research focuses on a new class of few-shot learning algorithm, which allows us to develop models with only 50 or 100 data points. »
Since its inception, Valence has collaborated with larger companies to help them develop their own drugs, especially in the field of oncology. For the future, Therence Bois, however, sees much bigger. “At the end of 2021, we made a round of financing to be able to expand our activities. What we are working on right now is to start developing our own drugs while continuing to provide access to our technology to other groups. »