Canadian researchers use AI to find a possible treatment for bacteria superbug

Read transcribed audio

Researchers have discovered a promising treatment for an antibiotic-resistant superbug — with the help of artificial intelligence.

Acinetobacter baumannii is a hospital-acquired pathogen that’s commonly found on surfaces in clinical settings. It can cause diseases such as pneumonia, meningitis and sepsis.

According to the World Health Organization, A. baumannii is a critical threat to patients whose care requires devices such as ventilators, due in large part to its resistance against most antibiotics.

“It’s remarkably challenging [to tackle],” said Jonathan Stokes, an assistant professor at McMaster University, in Hamilton, Ont., who led the research.

“When we go to search for new antibiotics, it necessitates that we start looking for chemicals, antibiotics that have brand new structures and brand new functions. You know, we have to develop a fundamentally new treatment,” he told The Current‘s Matt Galloway.

Usually, this involves testing hundreds of thousands of chemicals to see which ones work best against the disease. But Stokes says “that’s remarkably laborious and time-consuming and expensive.”

That’s why Stokes and the rest of the team, which included scientists at the Massachusetts Institute of Technology, turned to AI for assistance.

“Ideally, by leveraging these artificial intelligence algorithms, they can look at these chemicals much more rapidly,” he said. “And by looking at a broad array of chemicals very rapidly, they can help us prioritize which experiments to run in the laboratory.”

Stokes and his team published their findings in the journal Nature Chemical Biology on Thursday.

Training the model

Before the AI can find a chemical that could kill A. baumannii, Stokes and his team trained it by feeding it data on bacteria-killing chemicals and chemical structures “associated with the antibacterial activity that we want,” he said.

“We physically tested in the laboratory about 7,500 chemicals, looking at which ones inhibited the growth of Acinetobacter and which ones did not,” he said. 

Once the AI model was trained, the team could then show it new chemicals it had never seen before. It could then predict which of those chemicals it thought were antibacterial and which ones it thought weren’t.

Eventually, the AI discovered a new antibacterial compound they named abaucin. Further laboratory experiments found that it can treat A. baumannii-infected wounds in mice.

The next step, Stokes said, is to perfect the drug in the laboratory and then perform clinical trials.

This work highlights a promising lead in the fight against A. baumannii — and the role of AI technology in that cause.

“When we completed this project … I feel like we’re entering an era where AI approaches can meaningfully influence how we discover clinical medicine from the earliest stages of discovery,” Stokes said.

Large-scale experiments

For Stokes, AI promises to dramatically speed up scientific and medicinal research.

“Humans might not have to spend so much time and effort performing these large-scale experiments,” he said. 

WATCH: How AI could change the future of our health care

That promise resonates with other scientists, like Rahul Krishnan, an assistant professor in computational medicine at the University of Toronto.

“If it helps us get to discoveries even 10 per cent faster, that’s a huge win for society as a whole, because we can start making and discovering these drugs at a much faster scale,” he told Galloway.

My goal is to discover new antibiotics to save people’s lives. So if there are … powerful AI technological developments that help me achieve that goal, I am going to embrace them.-Jonathan Stokes

Krishnan, who studies the intersection of AI technology and health care, says the key idea for AI in medicine is to help clinicians make faster, safer decisions.

An AI could look at a patient’s medical records and use them “in conjunction with a predictive model to assist in clinical decision-making,” he said. For example, an AI could quickly predict whether a patient was likely to develop diabetes and then “have a clinician prescribe early interventions,” preventing more serious outcomes later on.

“From a public health standpoint, having the ability to have good predictive models deployed at scale might actually help individuals make better downstream decisions about their health,” he added.

Is AI data accurate, or ethical?

That’s not to say the introduction of AI wouldn’t have its challenges, though.

The growing popularity of AI in multiple fields has led to some warning it could lead to privacy and copyright violations and misinformation campaigns.

Executives, researchers and AI pioneers have warned that its unregulated use of AI could pose serious risks or even threats to humanity itself.

Krishnan says AI could be susceptible to biases that exist in the medical sphere, depending on the data used to train it.

“We know from a lot of studies that have been done over the decades that the health care system that we have in North America is incredibly, in some ways, unfair,” he said. 

“Those inequities are often translated into the data that are then fed into these algorithms. And if not corrected for at the point of training, these biases get encoded into the algorithm and every subsequent output that they put out.”

There’s also a risk of the AI making things up, even if it’s trained on reliable data.

“It, in some sense, can often hallucinate, and this is one of the failure modes of large language models … and obviously, that is a huge concern in the context of health care,” Krishnan said.

Stokes believes AI technology is advanced enough that it can be implemented now. But he says there’s still a lack of data “across many disease areas” to train these models.

“These AI models are … data hungry. They need to see a lot of examples in order to make robust predictions,” he said.

“So I think the acquisition of data with which we can train these models needs to be at the forefront of all of our thought.” 

Embracing AI in medicine

Krishnan sees a future where AI helps a clinician “automate away a lot of the simplistic cases,” freeing them up for more complex work.

“They can spend their cognitive effort and the cognitive cycles on the much more complex cases that demand their attention,” he said.

It’s this augmentation that leads to Stokes to believe that AI have a place in the laboratory and hospitals.

“My goal currently is to discover new antibiotics to save people’s lives,” he said.

“So if there are, you know, more robust, more powerful AI technological developments that help me achieve that goal, I am going to embrace them.”

---

Produced by Kate Cornick, Willow Smith and Magan Carthy.