Turbo-boosting pharmaceutical breakthroughs

The quest for new therapeutics is a multi-billion-dollar global industry. Pharmaceutical companies are hunting for drugs that work more effectively, that have fewer side effects or that target diseases that currently lack treatments. And one of the most urgent needs is finding new antibiotics, as growing antimicrobial resistance erodes the potency of current ones. 

But with most of the low-hanging pharmaceutical fruit already picked, that quest is getting harder. Traditional drug development has relied on either tweaking the formula of existing drugs or using automated techniques to screen large numbers of compounds for specific medicinal qualities. Today, those approaches are proving less and less productive. 

“It’s becoming much, much more expensive to find new drugs and bring those drugs to market,” says Ré Mansbach, a physicist at Concordia University and Canada Research Chair in Computational Physics/Biophysics. “To find things that are less obvious, we need to start thinking about new techniques.”

Grouping like with like: cataloguing medicinal molecules 

So Mansbach’s team is using artificial intelligence to categorize potential drugs. This novel approach will create strategic “search spaces” that promise to help in two crucial ways. 

First, the AI system will scour collections of chemical and biological compounds, sorting them according to their molecular sequences. Mansbach likens it to developing a library catalogue. Instead of grabbing a random book and hoping it’s the one you want, you can use the catalogue to zero in on the right shelf.

Figuring out which compounds should be grouped together isn’t simple. Which bits of their molecular structure do you focus on to decide if two potential drugs are similar?

However, once a “search space” has been categorized, AI makes it simple to pinpoint promising therapeutic candidates. “One thing that AI is really good at is looking at a lot of things in the blink of an eye,” Mansbach explains. “So you can narrow down the number of things that you have under consideration.”

AI can also help drug developers optimize for specific characteristics, such as the ability to penetrate cell membranes. Here, the challenge is finding the right balance between directing AI and giving it free rein. 

“Sometimes if you constrain your model too much, you’ll actually miss things,” Mansbach says. “It’s possible for the machine to pick up on a pattern that you weren’t aware of.”

Initially, their lab is focussing on antimicrobial peptides — short proteins with big potential to replace penicillin and other antibiotics that are losing their power. And thanks to CFI funding, they have the graphical processing units (GPUs) that speed up the AI-intensive analysis. 

Meanwhile, through Calcul Quebec — an academic resource-sharing initiative — other researchers have access to that computing power when their team isn’t using it, squeezing more value out of the equipment.

Accelerating the process of drug discovery

Ultimately, Mansbach hopes their findings will speed up drug discovery and cut pharmaceutical R&D costs. But it’s the opportunity of producing more fundamental insights that excite them the most. “We’re building competencies that I think will help commercial drug development in general. But we’re also building what I hope is a basic science understanding,” they say.

 

Finding new ways to find new drugs is really important for the health industry. It’s one of the big problems that we’re currently facing.

– Ré Mansbach, Concordia University

The research project featured in this story also benefits from funding from the Canada Research Chairs Program, the Digital Research Alliance of Canada, Mitacs and the Natural Sciences and Engineering Research Council of Canada.