Neural networks have helped scientists to develop an antibiotic capable of fighting a highly resistant superbug commonly found in hospitals.
The bug is called Acinetobacter baumannii and it is insidious.
“Acinetobacter can survive on hospital doorknobs and equipment for a week or longer, and can take up antibiotic resistance genes from its environment,” said Jonathan Stokes, an assistant professor of biochemistry and biomedical sciences at McMaster University. “It’s really common now to find A. baumannii isolates that are resistant to nearly every antibiotic.”
Stokes and his colleagues at McMaster University and MIT turned to AI to identify compounds that can combat the microbe. First, they exposed 7,500 different molecules to a strain of the bacteria grown in a lab dish to see whether they 1would inhibit its growth. They used that dataset to train a machine learning classifier to learn what chemical features in compounds gave the bacteria grief.
The model was then used to analyze a new dataset comprising 6,680 compounds it had not seen before, to predict whether they might make promising antibiotics.
The software identified hundreds of candidates in just two hours of runtime, and the researchers chose 240 for further experiments.
That process eventually produced nine candidate antibiotics, with a compound called “abaucin” found to be the most effective against A. baumannii.
Abaucin was previously studied as a potential diabetes drug. Now it’s tagged as a A. baumannii-hunter that selectively attacks the superbug.
Initial experiments with abaucin on mice showed it could suppress wound infections caused by the A. baumannii. The results were published in a Nature Chemical Biology paper on Thursday.
The researchers noted that abaucin isn’t as effective as conventional antibiotics, but because A. baumannii has developed resistance to common treatments, the compound identified by AI could represent a new class of antibiotics to target the bug.
“All of our experimental data suggests that abaucin inhibits a biological process in A baumannii called lipoprotein trafficking, which is an uncommon mechanism amongst current antibiotics used in the clinic,” Stokes told The Register. “We are currently focused on making structural analogs of abaucin to optimize its medicinal properties to maximize the chances that abaucin – or an analog of abaucin – could become a clinical antibiotic to fight A baumannii infections.”
He said the experiments demonstrate AI can be a powerful tool for drug discovery. “We can show these models vast numbers of chemicals and the models then tell us which chemicals have the property we care about. We can then focus our time and resources on experimenting on the most promising chemicals as suggested by the AI model. AI makes suggestions. Humans make decisions,” he told us.
James Collins, co-author of the study and a professor of medical engineering leading MIT’s Abdul Latif Jameel Clinic for Machine Learning in Health, agreed in a statement: “AI approaches to drug discovery are here to stay and will continue to be refined. We know algorithmic models work, now it’s a matter of widely adopting these methods to discover new antibiotics more efficiently and less expensively.” ®