Antibiotic to Kill Superbug Discovered Through AI

Scientists have discovered a potent and experimental antibiotic that will kill a deadly species of superbug — but they didnt discover it alone. They did so using artificial intelligence (AI) technology.

The researchers, based in the United States and Canada, said the technology allows them to accelerate the discovery of new drugs, and in the case of the new antibiotic, called abaucin, they used AI to narrow down several thousand possible chemicals to a small amount that could undergo laboratory testing, according to the BBC.

Infections that are increasingly resisting treatment with known antibiotics are proving deadly, with more than a million people a year expected to die from infections that can’t be treated.

The researchers in the AI case focused on the Acinetobacter baumannii bacteria, described as one of three superbugs the World Health Organization has determined as a “critical threat.”

According to the Centers for Disease Control and Prevention, the bacteria can cause infections in a patient’s blood, urinary tract, and wounds, or show up in the lungs, causing pneumonia. It can also live in a patient, or colonize, without causing infections or symptoms.

It can also survive on surfaces and medical equipment in hospitals and other care centers and is resistant to multiple antibiotics. 

Dr. Jonathan Stokes of McMaster University told the BBC that the bacteria is “public enemy No. 1” as there are commonly cases where it remains “resistant to nearly every antibiotic.”

Researchers using AI in the antibiotic discovery first had to train the technology to seek out the chemicals. They first sought out thousands of drugs and tested them manually on the Acinetobacter bacteria to see if it would either kill it or slow it down. Then they fed the information into the AI program to learn the chemical features of the drugs that could attack the superbug.

The results of the testing, published in Nature Chemical Biology, showed it took AI about an hour and a half to produce a short list. 

The researchers then tested 240 of the chemicals, finding nine potential antibiotics, including abaucin, which experiments showed could kill A. baumannii samples from patients and treat infected wounds in mice.

Stokes said, though, that the drug must be perfected in the laboratory and go through clinical trials, so he expects that the first AI-discovered antibiotics won’t be ready for prescriptions until at least 2030. 

This experimental antibiotic did not show effects on other species of bacteria and only worked on A. baumannii. 

But still, Stokes said, AI not only increases the rate of tests but also decreases the cost of discovering new classes of antibiotics that are needed.