Researchers turn to AI for help fighting superbugs

This story is part of CBC Health's Second Opinion, a weekly analysis of health and medical science news emailed to subscribers on Saturday mornings. If you haven't subscribed yet, you can do that by clicking here.

Vulnerable patients with superbugs need antibiotics to work better, say doctors and scientists who are encouraged by an AI tool that invented new chemicals to treat drug-resistant gonorrhea and MRSA.

In the arms race against drug-resistant superbugs, humans have been falling behind. Infections from antibiotic-resistant strains kill more than 1.2 million people a year globally, leading to an "urgent global health threat," according to the World Health Organization.

That means doctors need new ways to get the upper hand on impossible and hard-to-treat infections. There have been tweaks to existing classes of antibiotics, but no new major antibiotic has been discovered in nearly 40 years.

Now, medical experts say an AI tool has creatively designed new chemicals that might be able to treat major superbugs.

Last month, medical engineers based at Massachusetts Institute of Technology (MIT) reported in the journal Cell that they used a generative AI model to suggest unique compounds with antibacterial potential against two scourges: the drug-resistant sexually transmitted infection gonorrhea and MRSA, which stands for methicillin-resistant Staphylococcus aureus, a common cause of hospital infections in Canadian hospitals.

While the new compounds still need to go through clinical trials to check for safety and efficacy and be approved by regulators like Health Canada to be used by patients, they have shown potential against the drug-resistant strains in laboratory tests on mice.

"These computational models elegantly and precisely design compounds from scratch," with antibiotic properties, said Akhila Kosaraju, CEO and president of Phare Bio, a nonprofit biotech, who partners with the MIT team. "This is the true, extraordinary breakthrough here."

Treating infections quickly is crucial for many conditions, from complicated pregnancies and organ transplants to advanced cancer cases, said Romney Humphries, a professor and division director for laboratory medicine at Vanderbilt University Medical Centre in Nashville, Tenn.

"We have patients who have infections with no available antibiotic," said Humphries. "This is really scary."

Kosaraju said the scale of antibiotic resistance is massive and needs new approaches that an individual clinician can't solve. Phare Bio, the nonprofit biotech she leads, is funded by philanthropists with the goal of bringing 15 novel antibiotic candidates to the early-stage research pipeline in five years.

"As a physician by training, I know you only have to spend a few days in the hospital system, whether it's in the U.S. or Canada or elsewhere, to see the catastrophic toll of antibiotic resistance," she said.

Previously, scientists would select a thousand compounds from a database and then test in a lab whether they killed bacteria like E. coli, with yes or no responses, Kosaraju said.

Generative AI is a type of artificial intelligence that uses pattern-matching technologies to learn from large datasets and create content in response to prompts.

In the MIT study, the generative AI model learned which chemical structures kill bacteria and identified new possibilities. The top candidates against gonorrhea and MRSA were synthesized and successfully tested in the lab and mouse models of the infection.

"Our thesis is that with AI and now generative AI, we will be taking fewer but better shots on goal," Kosaraju said.

WATCH | Family of woman who died of septic shock after childbirth warns of signs:

A GTA family is calling for accountability and new hospital policies after a mother died earlier this summer, following childbirth. As CBC's Sarah MacMillan explains, the family says staff at Mississauga's Credit Valley Hospital didn't act quickly enough to recognize and treat sepsis.

The study's results matter because the current approach of throwing everything at the bacteria to see what might work isn't always successful, said Humphries, who was not involved in the paper.

There are a few different ways to kill or control bacteria — like making the cells explode, starving them, or preventing the microorganism from making copies of itself. In the study, the chemicals found appeared to be able to weaken an important bacterial membrane on gonorrhea and MRSA.

Next, Phare Bio is working on versions of these compounds that meet the criteria patients need, like being able to be manufactured as a pill that can be taken orally at home instead of an injection in hospital.

Beyond the scientific challenge superbugs pose, there's the money problem. It takes at least 10 years and hundreds of millions of dollars to get a new pharmaceutical for patients. Antibiotics face a deeper chasm of profitability than other medications, Kosaraju said.

This week, the Public Health Agency of Canada published the country's new list of priority microbes. From a starting list of 155 pathogens, they flagged 29 as "significant risks to Canadians" based on incidence, treatability, transmission and health equity.

Drug-resistant gonorrhea and carbapenem-resistant enterobacterales, which can cause urinary tract and kidney infections, sepsis and meningitis when a person's immune system is weakened, topped the federal list.

But coming up with new potential antibiotics the old-fashioned way of testing in a Petri dish is like finding a needle in a haystack, said Eric Brown, a professor in biochemistry and biomedical sciences at McMaster University in Hamilton, Ont., who also uses generative AI in the hunt against superbugs.

Bacteria have up to 4,000 genes, Brown said. Currently, scientists only understand a couple dozen pairs of those genes, never mind how the bacteria interacts with humans. Without knowing how the bacterial genes work with each other and against people, new potential antibiotics are unlikely to work.

"It's a little bit like trying to predict the weather," Brown said. "It's a complex system that is very challenging to get your head around. It turns out that AI and other kinds of math can help us."

Solving antibiotic resistance also needs a combination of biology, chemistry, physics, computer science and programming, as well as statistics, to analyze and interpret the clues AI offers.

While it's important to have new potential antibiotic compounds like MIT's that look promising, that's just step one, Humphries cautioned.

"There's a really big steep curve to come over after you've done that step to say, OK, but is it safe in people?" Humphries said. "Does it stay in your body long enough if we give it to someone that it can actually be effective against the bacteria?"