The artificial intelligence that discovered a new antibiotic (in two hours)

An artificial intelligence developed by MIT has given wings to the search for an antibiotic against one of the most fearsome hospital superbugs, theAcinetobacter Baumannii. A specially trained machine learning program has scanned thousands of potential drugs until it has identified a new antibiotic capable of neutralizing the microorganism, which concerns its ability to develop resistance to the classes of drugs already used to fight it. The discovery was published on Nature Chemical Biology.

A tough nut to crack. L’A. baumannii it is a widespread bacterium in hospitals responsible for even very serious infections, such as pneumonia and meningitis. It is capable of resisting on surfaces (such as door handles or medical instruments) even for a month, and can acquire antibiotic-resistance genes from the environment in which it proliferates: for these reasons it often happens to find resistant samples of the bacterium to virtually any currently available antibiotic.

Phase 1: learning. A team of scientists from the Massachusetts Institute of Technology (MIT) in Cambridge and McMaster University in Hamilton, Canada, thought of exploiting AI for a complex task such as neutralizing this “public enemy” of hospital infections. To educate the computational model to use in the study, the scientists exposed samples of A. baumannii grown in the laboratory to 7,500 chemical compounds known and already used in the pharmaceutical field, so as to show the algorithm which molecules were more effective in inhibiting the growth of the bacterium. Through this training, the AI ​​”absorbed” the characteristics to look for to defeat the microorganism.

Stage 2: research. Armed with this treasure trove of instructions, the AI ​​analyzed a database of 6,680 compounds it did not yet “know” about, already approved by the US FDA and cataloged in the Broad Institute’s Drug Repurposing Hub (an archive that encourages repurposing, or reuse, of old drugs developed for other purposes, shortening research times). In less than two hours, the model highlighted 240 promising drugs against the superbug that the scientists tested in the laboratory, focusing on those with very different characteristics from those used so far, to reduce the risk of resistance phenomena.

Nine of these were antibiotics and one, renamed abaucin and initially studied as a potential drug against diabetes, it has proved to be particularly potent againstA. baumannii – but not against other species of antibiotic-resistant bacteria, such as lo Pseudomonas aeruginosalo Staphylococcus aureusand the family of Enterobacteriaceae.

One goal. This ability to only neutralize the desired superbug The researchers liked it a lot, because it reduces the risk of a rapid spread of resistance episodes against the new antibiotic.

Another advantage is that the drug could act in a targeted way, without also eliminating the beneficial bacteria that colonize the human intestine – and perhaps pave the way for other infections as a result.

Unedited intervention. The mechanism of action of abaucin is also rather unique: the new antibiotic subtracts energy from theA. Baumani by interfering with a process that helps the passage of proteins from inside a cell to the cell walls. The hypothesis is that theA. baumannii performs this task in a slightly different way than other gram-negative bacteria (the type of bacteria to which the pathogen belongs) and that this exclusivity favors the selective efficacy of the antibiotic.

Phase 3: Action! For now abaucin has proven effective in healing wounds infected by A. baumannii in mice, and it also worked against samples of the bacterium resistant to several antibiotics isolated from human patients. Now the team is working to optimize the medical properties of the molecule so that it can be used as an antibiotic. Meanwhile, he plans to use AI again to look for new solutions against other superbugs, like lo Staphylococcus aureus and it Pseudomonas aeruginosa.