MIT researchers use artificial intelligence to identify powerful new antibiotic

Shawn Knight

TechSpot Staff
Staff member

Only a small number of new antibiotics have been developed over the past few decades due to the cost and time associated with screening them. In fact, most newly approved antibiotics are simply minor variants of existing drugs. But now thanks to the power of artificial intelligence and the growing need to fend off antibiotic-resistant bacteria, that is changing.

Researchers at MIT created a machine learning model to look for chemical features that make molecules effective at killing E. coli, training it on around 1,700 FDA-approved drugs and 800 natural products. Once trained, the model was tested on the Broad Institute’s Drug Repurposing Hub which consists of around 6,000 compounds.

The model honed in on a molecule that it predicted to have strong antibacterial properties and a chemical structure that differed from existing antibiotics. A separate machine learning model indicated it might also have low toxicity to human cells.

The molecule, which they dubbed halicin after the computer “HAL” in 2001: A Space Odyssey, was tested against dozens of strains of bacteria and found to be effective in eradicating many of them including Acinetobacter baumannii, Clostridium difficile and Mycobacterium tuberculosis.

In mice infected with a particular strain of A. baumannii that is resistant to all known antibiotics, halicin completely cleared up infections within 24 hours.

Researchers believe halicin kills bacteria by rattling its ability to maintain an electrochemical connection across cell membranes. This gradient is necessary to produce ATP and without it, the cells die. What’s more, it proved effective against mutations.

Over a 30-day period, E.coli didn’t develop any resistance to halicin. In comparison, the bacteria developed resistance to the antibiotic ciprofloxacin within one to three days. By day 30, the bacteria was roughly 200 times more resistant to ciprofloxacin than it was initially.

The researchers’ model has since been used to identify 23 other candidates, two of which were particularly powerful against bacteria. They plan to conduct additional testing on halicin and work with a nonprofit or pharmaceutical company to develop it for human use.

Masthead credit: Antibiotics by ESB Professional.

Permalink to story.


Uncle Al

TS Evangelist
This is no less amazing that the first mold based antibiotics often called "wonder drugs". With any luck they will be able to further develop some that do not increase resistance of known strains of pathogens or similar microbials. Looks like AI has certainly put a feather in it's cap with this one!
Last edited:


TS Evangelist
And how long will it take for this to come to market, not to mention the price.
Then, in 2-3 years they will find it creates some odd ball problem, and the "we sue for anything"
lawyers will have commercials running 24/7 on television, radio & the internet. :(
This sounds like it could be promising, and with any luck, bacteria will not evolve to make the treatment ineffective. has some projects along these lines. Anyone who would like to may contribute computing power to the effort.
Don't worry, they will. Human greed and stupidity is risking to render one of medicine's greatest gifts ineffective.

That said, using AI to discover new drugs is a great thing. Your suggestion of distributed computing as well.
  • Like
Reactions: Dimitrios


TS Booster
Let's use that AI to resolve AIDs, cancer, and other unsolved mysteries. We resolve all of that... we won't have much for population control. We need to start colonizing Mars soon. I think I'm seeing the pattern right now. Things are going to be weird when I become a senior citizen.


TS Evangelist
The biggest issue with antibiotics is the blanket way they kill all bacteria-good and bad within the body. To be able to make an antibiotic that can specifically target the offending ones while leaving the rest unscathed would be an enormous breakthrough. Even if it was just able to weaken the bad bacteria enough so as to allow the body and the good bacteria to win the war.


TS Evangelist
This is why we need proper quantum computers with millions of qubits. Creating molecules requires too much time and processing power for regular computers.