2024-01-02 18:00:06
Discovering new classes of antibiotics using artificial intelligence (AI) is becoming a credible prospect. Enough to clear a horizon darkened by the scourge of growing antibiotic resistance. A team from the Broad Institute of MIT and Harvard (Cambridge, Massachusetts) has identified a class of compounds capable of killing drug-resistant bacteria. Their measured toxicity for human cells appears very low, which makes it possible to see them as drug candidates. But we are still far from the development of an antibiotic.
The interest of this research, which was the subject ofa publication, on December 20, in Nature, lies above all in the method used, more than in the molecules identified. It is no coincidence that Felix Wong, the first signatory of the article, is a physicist and mathematician, not a microbiologist. Biomedical research already uses AI, but here it is the deep learning mode imagined that is innovative.
The researchers first determined the antibiotic activities of 39,312 compounds, the chemical substructures of already known molecules, as well as their cytotoxicity profiles on human cells. Neural networks were thus trained to identify chemical structures associated with antimicrobial activity. They were then used to screen more than 12 million compounds and predict their antibiotic activity and cytotoxicity.
Read also: Article reserved for our subscribers “Against antibiotic resistance, public research and production of medicines are essential”
Ultimately, 283 compounds were selected to be empirically tested against Staphylococcus aureus. Tests on mice identified a class of structure that was antimicrobial against methicillin-resistant staphylococcus and enterococci resistant to vancomycin, two standard antibiotics. The word is out, this would be the path for a new “class” of antibiotics. For several decades, the antibiotics arriving on the market have only been improved or extended-spectrum versions of existing families.
“A spectacular result”
Didier Mazel, head of the Bacterial Genome Plasticity unit at the Pasteur Institute, welcomes, in this study, a “spectacular result”. “They unmasked a family of molecules that were not known for their antimicrobial activity, this validates their model”, adds the geneticist. If the use of AI has already made it possible, for example, to obtain new protein structures of possible biological interest, the computing power to which the American team had access has made it possible to take a step forward.
You have 45% of this article left to read. The rest is reserved for subscribers.
#class #molecules #unearthed #artificial #intelligence