Scientists use AI to discover antibiotic for 'very difficult to treat' bacteria
With the use of artificial intelligence, scientists in Canada and the United States have discovered an antibiotic that could be used to fight a deadly, drug-resistant pathogen — and they hope to employ a similar process to discover treatments for other challenging bacteria.
In a study published in the journal Nature Chemical Biology Thursday, the researchers at McMaster University and the Massachusetts Institute of Technology shared their promising findings about the new antibacterial treatment, which they named abaucin.
Jon Stokes, a lead author of the research paper, said the antibiotic could be used to fight Acinetobacter baumannii, which the World Health Organization has identified as one of the world’s most dangerous antibiotic-resistant bacteria.
“In my opinion, it is public enemy No. 1 for antibiotic resistance — it’s very difficult to treat,” said Stokes, who is an assistant professor in the department of biochemistry and biomedical sciences at McMaster.
“It tends to live in hospital settings, so you find it on doorknobs and hospital equipment and stuff. And it's really challenging to sterilize, so it can survive on these hospital surfaces for prolonged periods of time.”
The bacterial pathogen, also known as A. baumannii, can cause pneumonia, meningitis and infect wounds, all of which can lead to death. It’s also able to pick up DNA from other species of bacteria in its environment, which can encode antibiotic-resistance genes, Stokes noted.
In order to discover an antibiotic to fight the highly drug-resistant pathogen, Stokes said the researchers tested roughly 7,500 molecules with different structures in a lab to see which of them were able to inhibit the growth of A. baumannii and which of them were not.
Then, he said they trained an AI model to understand what chemical features result in molecules that have A. baumannii activity.
“Once we trained our model, we could then start showing the model a bunch of pictures of brand new molecules that it had never seen, flash card style,” Stokes explained.
“And then, based on what the model learned during training, it would predict which chemicals it thought were antibacterial, and which ones it thought were not.”
After that, the researchers acquired the molecules that the AI model predicted were antibacterial and tested them to see how well they could fight off A. baumannii.
“And that was easy, because instead of having to test thousands of molecules, we were testing a couple hundred,” Stokes said.
“We ended up finding this one molecule that was potent at inhibiting the growth of Acinetobacter in the laboratory — and it was structurally unique relative to every other known antibiotic we have. So this AI model helped us rather efficiently pull out an interesting molecule with antibacterial properties against the bug we were trying to kill.”
Stokes conducted the research with James J. Collins, a professor of medical engineering and science at MIT, McMaster graduate students Gary Liu and Denise Catacutan, as well as Khushi Rathod, a recent McMaster graduate.
Stokes said their research offers proof that the application of AI methods can “meaningfully influence” the discovery of new antibiotics across a whole bunch of different challenging pathogens. And he hopes to use similar methods to discover other antibacterial treatments.
“I'm not saying that AI is a panacea — it's not going to solve all of our problems for us — but it's a very powerful tool in our toolbox with which we use to find new medicines for people.”