For the first time, AI has helped scientists identify a rare antibiotic that could combat drug-resistant infections. A team of researchers from the Massachusetts Institute of Technology (MIT) and McMaster University used machine learning, a branch of Artificial Intelligence (AI), to sift through millions of chemical compounds in search of one that would kill dangerous bacteria. The resulting antibiotic, dubbed halicin after the sentient computer from Stanley Kubrick’s film 2001: A Space Odyssey, is both simple and effective, and its discovery demonstrates how AI can be used to speed up research and development.
Researchers from MIT’s Abdul Latif Jameel Clinic for Machine Learning in Health worked with researchers at McMaster University to screen 7,500 chemicals and find those that inhibited the growth of a particular type of dangerous bacteria, Acinetobacter baumannii. These bacteria thrive in hospitals, where they can survive on doorknobs and equipment for long periods and take up antibiotic resistance genes from the environment. Acinetobacter-associated infections are responsible for more than 300,000 deaths a year worldwide. The researchers’ goal was to find a compound that could target Acinetobacter without inhibiting common bacteria, which often cause drug-resistant infections. The researchers’ screening process ultimately yielded a compound they named again, and it showed promise in tests on mice with wound infections caused by Acinetobacter.
In a paper published Thursday in Nature Chemical Biology, the researchers explained how they used their AI system to find the antibiotic. The team began by training their neural network on 7,500 known molecules that either do or don’t fight Acinetobacter, so the computer program knew what to look for.
Then, they asked the neural net to search through a database of 6,000 chemicals at various stages in clinical trials for those that matched what it had learned about E. coli’s structure. The computer found halicin, which resembles the structure of other antibiotics but is otherwise different.
Once they verified the identification, they took a closer look at the compound and found that it did indeed suppress Acinetobacter’s ability to grow. However, it also did not appear to impact healthy cells, meaning it could be used as a treatment in the hospital without causing side effects.
The study’s authors believe that the success of their work will open the door to further exploration of AI-based methods for identifying new antibiotics, and they plan on continuing their research with other drug targets, including multidrug-resistant bacteria. They also hope to explore ways to improve the accuracy and reliability of their model to ensure it can be used as a tool for medical discoveries.