A powerful antibiotic discovered thanks to the artificial intelligence

A powerful antibiotic was discovered for the very first time thanks to machine learning (literally “machine learning”), a field of study of artificial intelligence which is based on mathematical and statistical approaches to give computers the ability to learn from data, that is, to improve their performance in solving tasks without being explicitly programmed for them.

Now, thanks to artificial intelligence , a team from the Massachusetts Institute of Technology (MIT) claims that halicin (the powerful antibiotic in question) kills some of the most dangerous strains of drug-resistant bacteria in the world.

This drug works differently from existing antibacterials and is the first of its kind to be discovered by AI browsing large digital libraries of pharmaceutical compounds.



Tests by researchers have shown that the drug successfully eliminates a range of antibiotic-resistant bacteria strains, including Acinetobacter baumannii and Enterobacteriaceae, two of the three pathogens to be given high priority and that the World Health Organization Health (WHO) also classifies as "critical".

The culture plate on the right contains bacteria resistant to all the antibiotics tested so far. Credit: Science History Images / Alamy

"In terms of antibiotic discovery, this is absolutely a first," said Regina Barzilay, project lead researcher and machine learning specialist at MIT. "I think it's one of the most powerful antibiotics discovered to date," added James Collins, a bioengineer on the MIT team. "It has remarkable activity against a wide range of pathogens resistant to current antibiotics."

Be aware that antibiotic resistance occurs when bacteria mutate and evolve to bypass the mechanisms that antimicrobial drugs use to kill them. Experts say that without new antibiotics to fight resistance, 10 million lives around the world could be threatened by infections each year by 2050.

AI to discover potential new antibiotics

In order to discover new antibiotics, the researchers first trained a “deep learning” algorithm to identify the types of molecules that kill bacteria. To do this, Scientists trained it to analyse the structure of 2,500 drugs and other compounds to find those with the most anti-bacterial qualities that could kill E. coli.

Once the algorithm learned which molecular features make good antibiotics, scientists let it browse a library of more than 6,000 compounds being studied to treat various human diseases.

And, rather than looking for potential antimicrobials, the algorithm focused on compounds that seemed effective, but were different from existing antibiotics. Therefore increased the chances that these drugs will act in a radically new way and that bacteria have not yet developed resistance to them.

Jonathan Stokes, the first author of the study, said it took a matter of hours for the algorithm to assess the compounds and come up with some promising antibiotics. One, which the researchers named “halicin” after Hal, the astronaut-bothering AI in the film 2001: A Space Odyssey, looked particularly potent.

Very promising positive results!

Since this discovery, researchers have been able to treat many drug-resistant infections with halicin, a compound that was originally developed to treat diabetes, but which ultimately did not work. Tests on bacteria collected from patients have shown that halicin can eradicate Mycobacterium tuberculosis , the bacteria responsible for tuberculosis, as well as strains of enterobacteriaceae resistant to carbapenems, a group of antibiotics considered to be the last resort for such infections.

Halicin has also successfully eliminated difficult infections and multidrug-resistant Acinetobacter baumannii infections in mice. To research new drugs, the team then turned to a large digital database of around 1.5 billion compounds. They adjusted the algorithm so that the latter analyzes 107 million of these compounds. Then, three days later, the program returned a shortlist of 23 potential antibiotics, two of which appear to be particularly potent.

Use AI to find other more targeted antibiotics

Now scientists plan to search more databases for potential antibiotics. Stokes stated that it would have been impossible to screen all of the compounds by conventional means of obtaining or manufacturing the substances, and then to test them in the laboratory. "Being able to perform these experiments on a computer greatly reduces the time and cost of examining these compounds," he said.

Barzilay now wants to use the algorithm to find more selective antibiotics in the bacteria they kill. This would mean that taking the antibiotic would kill only the bacteria causing an infection, and not also all the healthy bacteria that live in the gut in particular.

Even more ambitious, scientists aim to use the algorithm to design powerful new antibiotics from scratch. "This work is truly remarkable," said Jacob Durrant, who works on computer-aided drug design at the University of Pittsburgh. “Their approach highlights the power of computer-aided drug discovery. It would be impossible to physically test 100 million compounds to determine antibiotic activity.”



Indeed, if we take into account all the typical costs of drug development, in terms of time and money, any method that can accelerate discovery, as is the case here thanks to artificial intelligence, has the potential to have a very significant impact.

Bibliography: A Deep Learning Approach to Antibiotic Discovery Jonathan M. Stokes, Kevin YangKyle Swanson, Wengong Jin, Andres Cubillos-Ruiz, Nina M. Donghia, Craig R. MacNair, Shawn French, Lindsey A. Carfrae, Zohar Bloom-Ackerman, Victoria M. Tran, Anush Chiappino-Pepe, Ahmed H. Badran, Ian W. Andrews, Emma J. Chory, George M. Church, Eric D. Brown, Tommi S. Jaakkola, Regina Barzilay, James J. Collins VOLUME 180, ISSUE 4, P688-702.E13, https://doi.org/10.1016/j.cell.2020.01.021