“Our ability to control infection upholds all of modern medicine,” says Jon Stokes, professor of biochemistry and biomedical sciences at McMaster University.
His statement follows a newly published study on AI-driven antibiotic discovery where he and his colleagues at Massachusetts Institute of Technology (MIT) developed an artificial intelligence algorithm to help identify a possible new antibiotic.
“Without effective antibiotics, we are unable to safely perform invasive surgeries, conduct organ transplants, and provide cancer chemotherapy treatments,” Stokes tells Happiest Health.
In the study published in Nature Chemical Biology, scientists used AI-driven antibiotic discovery to identify an antibiotic candidate, with significant time savings in the process.
They set their sights on Acinetobacter baumannii, a drug-resistant bacteria commonly found in hospitals and causative agent of pneumonia and meningitis.
“Acinetobacter can survive on hospital doorknobs and equipment for long periods of time, and it can take up antibiotic resistance genes from its environment. It’s common now to find A. baumannii isolates that are resistant to nearly every antibiotic,” says Stokes, a former MIT postdoc researcher.
Also read: Bacteriophage therapy: an age-old answer to combat antibiotic resistance
Tackling antibiotic resistance
There is a growing need for more narrow spectrum antibiotics that can have a more targeted approach against drug resistant bacteria, as current antibiotics are slowly becoming redundant against bacteria that constantly adapt.
“Given that we are facing an antibiotic resistance crisis, it is essential to discover novel antibiotics to treat bacterial infections that are resistant to current antibiotics,” says Stokes.
He adds that despite new antibiotic discoveries it is very likely that what is implemented in the clinic will eventually become obsolete as bacteria are highly dynamic organisms.
Calling them ‘survival experts’ he says, “It is our job to continually discover new antibiotics to ensure we can properly treat bacterial infections in the future.”
But finding something that works take a lot of time and capital, a challenge that could be overcome using AI-driven antibiotic discovery.
Using AI in drug discovery
For human researchers, screening a chemical collection of over 10,000 molecules — which includes involves figuring out which of them could inhibit the growth of the bacteria — could take weeks to complete.
“Using a trained AI algorithm, the same 10,000 molecules could be assessed for antibacterial activity within a couple hours,” says Stokes.
And this is exactly what they did when screening 7,000 potential compounds from an existing library to see which one could potentially inhibit the growth of A. baumannii.
The analysis, completed in under two hours, identified several hundred promising compounds out of which 240 were selected for experimental testing in the lab. The focus was on compounds with unique structures that differed from existing antibiotics.
Studies in mice
Once the AI identified a feasible compound, abaucin, they tested its effects on mice that had wound infections caused by A. baumannii as well as on drug-resistant of the same bacteria isolated from humans.
They found that abaucin was able to kill bacterial cells by affecting a process called lipoprotein traffic which cells use to transport proteins in the cell.
However, A. baumannii does this process a bit differently than other strains of bacteria making abaucin a selective antibiotic.
This “narrow spectrum” killing ability of the antibiotic makes it a perfect candidate for infections and minimizes the risk of bacteria rapidly spreading resistance against the drug. It also spares the beneficial bacteria in the gut who are killed because of broad-spectrum antibiotics we use.
Future directions
“A lot of work is required to turn a hit molecule from a screen into a clinical antibiotic,” says Stoke.
The next steps will include checking if the AI-identified compound is toxic in human cell lines and in animal models.
“We must ensure that the molecule will get to the site of infection and remain there long enough at high enough concentrations to eradicate the pathogenic bacteria,” adds Stokes.
The existing process of drug discovery can often take over ten years for a single drug to reach market. Can AI simplify the process? Stoke thinks so, adding that AI models could help in cutting this time span down.
“From early hit discovery to clinical optimization, [AI] will serve to both increase the speed to getting new drugs to patients, as well as decrease the costs, thereby making new drugs more equitably accessible by everyone who needs them,” concludes Stokes.
