A team of researchers, including several from McMaster University, used artificial intelligence to discover a promising new drug for a notoriously hard-to-treat, antibiotic-resistant “superbug.”
According to the study, published Thursday in the journal Nature Chemical Biology, the lab’s AI was able to comb through thousands of molecules in mere hours to hone in on the drug — results the researchers say could revolutionize antibiotic drug discovery.
Currently, humans are in a losing arms race against bacteria — the germs are evolving resistance to antibiotics at an alarming rate, far faster than new drugs could be developed. With advancements in AI and technology, however, humanity may eventually accelerate drug discovery by so much that bacteria can’t keep up, one of the study’s authors said.
Jonathan Stokes, an assistant professor of biochemistry at McMaster who led the study, told the Star his superbug target, a bacterium named Acinetobacter baumannii, is among the most dangerous and difficult-to-treat drug-resistant germs in the world.
“Acinetobacter is one of the most, if not the most, urgent bacterial pathogens for which new antibiotics are required,” Stokes continued. “That’s the logic behind why we embarked on this project.”
AI is turbocharging drug discovery
AI shows a remarkable ability to analyze vast quantities of information in a fraction of the time it would take a human. Stokes’s team leveraged this power to algorithmically screen nearly 7,000 chemicals — a “relatively small set of chemicals” for AI — to see if any proved effective against the infection.
To his team’s surprise and delight, not only did they hit on a candidate in hours — a process that would’ve previously taken weeks — the drug even showed a rare preference for acinetobacter while ignoring other, potentially helpful, bacteria.
“It was very exciting when we started observing that it wasn’t working against a whole bunch of other pathogens,” Stokes said. “That observation suggested it was doing something in Acinetobacter that was quite different from other antibiotics — and that’s what we were looking for, novel structures that have novel function.”
The drug, which the team named abaucin, appeared so promising the researchers stopped scanning additional molecules: “We found something that was worth a lot of time and money invested,” Stokes said. In reality, their program is able to comb through hundreds of millions of potential drugs in a matter of weeks — a feat impossible to match through traditional methods, he added.
To achieve their results, they first had to train the AI on 7,500 known chemicals and their interactions with Acinetobacter, so the computer program knows what works and what doesn’t.
AI antibiotic shows promise
After identifying the drug, Stokes’s team then tested abaucin on mice who had their wounds infected with Acinetobacter — a common way the bug spreads.
“What we observed specifically was abaucin was able to suppress the severity of infection” compared to mice given current antibiotics or no treatment — promising results at this early stage, he said.
Now Stokes’s lab is working on tweaking abaucin to improve its potency and other medicinal qualities, in hopes it would eventually make it to clinics. “It’s a long road between now and clinical trials,” he said. “There are still many things about abaucin that we want to optimize and improve.”
Justin Nodwell, a professor of biochemistry at the University of Toronto who is unaffiliated with the study, found its results “really interesting.”
“Finding a novel antibiotic against something like Acinetobacter, which is notoriously resistant, is a big deal,” he said — especially one that ignores other bacteria in the vicinity, a “terrible side effect” of most modern antibiotics.
He was impressed with the team’s method, calling it a “harbinger for what’s coming in the future.”
How AI could change medicine
Traditionally, discovering an antibiotic takes about a decade and costs “easily more than a billion dollars,” Nodwell said. AI is expected to reduce costs both in time and finances, though it’s yet unclear to what extent, he continued.
According to Phillip Kim, a professor at U of T’s Donnelly Centre, much of drug discovery consists of “throwing spaghetti at the wall” and seeing what sticks.
“You’re screening many, many, many compounds to find one that does what you need it to do, then you spend many, many, many years tweaking it to make it better” Kim, who is unaffiliated with the study, told the Star. “The big promise of AI is you can do all of that on the computer.”
Not only would this drastically speed up the process, it would allow AI to scan far more compounds than humans, Kim said — increasing the chance of finding an ideal molecule that does everything we want it to do.
“The big hope is that machine learning methods are becoming powerful enough that the (drug discovery) process is going to get much more efficient, much cheaper and much, much more accurate,” he said.
That overhaul “hasn’t happened yet — but there is a lot of evidence academically and in the industry that it will happen.”
JOIN THE CONVERSATION
does not endorse these opinions.