The move is the latest example of how artificial intelligence tools will become a powerful force in transforming science and medicine.

Antibiotics can kill bacteria, but there has been a lack of new drugs for decades, and the development of resistance to these drugs has made treating bacterial infections more difficult.

Statistics indicate that more than one million people die annually from infections caused by bacteria that are resistant to antibiotic treatment.

The researchers focused on one of the more complex bacteria, called Acinetobacter baumannii, which can infect wounds and cause pneumonia.

You may not have heard of them, but they are one of three types of superbugs that the World Health Organization has described as a “severe” threat.

This bacteria is often resistant to many antibiotics, which is a problem in hospitals and health care homes because it can live on surfaces and medical equipment.

Jonathan Stokes, a doctor at McMaster University, describes bacteria as “the most common enemy” because it is “already common” to find cases where they are “resistant to every antibiotic”.

To find a new antibiotic, the researchers first trained artificial intelligence, using thousands of drugs known for their precise chemical composition, and manually tested them on Acinetobacter baumannii to see which species could slow down or kill the bacterium.

The researchers feed this information to the AI ​​so that it can learn the chemical properties of drugs that can attack disease-causing bacteria.

Next, the researchers let the AI ​​work on a list of 6,680 chemical compounds of unknown potency. The results, published in the journal Nature Chemical Biology, showed that the AI ​​took about an hour and a half to select a short list.

The researchers tested 240 species in the lab and came up with nine potential antibiotics, including the most powerful antibiotic, “Apaucin.”

Laboratory tests showed that this antibiotic treated infected wounds in mice and was able to kill specimens of the superbug “A. baumannii” in some patients.

Despite this, Stokes said, “It’s time to start.”

The next step is to develop a drug in the laboratory and then conduct clinical trials with it. It is expected that it will take until 2030 for the first antibiotics to be prescribed to patients with the help of artificial intelligence.

It is strange that this test antibiotic had no effect on other types of bacteria because it was only effective against the bacterial type “A. Baumannii”.

Many antibiotics kill bacteria indiscriminately, and the researchers believe that the precision of the antibiotic “Abucin” will make it harder for bacteria to develop resistance with fewer side effects.

AI can, in principle, screen tens of thousands of potential chemical compounds, which would be difficult to do manually in practice.

“Artificial intelligence increases the rates (of effectiveness) and lowers the cost, so we can find these new classes of antibiotics,” Stokes said.

Researchers tested the principles of AI-assisted antibiotic discovery on coliform bacteria in 2020, but have now applied that knowledge to focus on more dangerous organisms. They also plan to conduct further research on Staphylococcus aureus and Pseudomonas aeruginosa.

Professor James Collins from the Massachusetts Institute of Technology said: “This result supports the hypothesis that artificial intelligence can accelerate and improve the scope of our research into new antibiotics.”

“I’m excited that this work shows that AI can be used to fight pathogens like A. baumannii,” he added.