Massachusetts Institute of Technology (MIT) researchers have identified a new class of antibiotic compounds that can kill methicillin-resistant Staphylococcus aureus (MRSA).
Responsible for over 10,000 deaths a year in the US, MRSA is a drug-resistant bacterium that can lead to deadly infections, including skin infections, pneumonia, and sepsis.
Researchers trained a type of artificial intelligence (AI) known as deep learning and generated training data after testing around 39,000 compounds for antibiotic activity against MRSA.
Deep learning models identify chemical structures that are associated with antimicrobial activity, sifting through millions of other compounds to generate predictions of which ones have strong antimicrobial activity.
This data was then fed to the model, as well as information on the chemical structures of the compounds.
Furthermore, the identified compounds also showed significantly low toxicity against human cells, meaning that they are strong drug candidates.
Using a search algorithm known as Monte Carlo tree search, researchers were able to see how the model generated an estimate of each molecule’s antimicrobial activity and its predictions, which could lead to additional drugs that could potentially work better than those identified by the model.
After training three additional deep learning models to predict which compounds were toxic to three types of human cells, researchers learned that compounds could kill microbes with minimal adverse effects on the human body.
The models identified compounds from five different classes that were predicted to be active against MRSA after researchers screened around 12 million compounds that were commercially available.
In a lab dish, they tested around 280 compounds against MRSA to reveal two compounds from the same class as promising antibiotic candidates.
Additionally, using two mouse models, one of MRSA skin infection and one of MRSA systemic infection, each compound reduced the MRSA population by a factor of ten.
Felix Wong, postdoc at MIT’s Institute for Medical Engineering and Science and the Broad Institute of MIT and Harvard, said: “The molecules are attacking bacterial cell membranes selectively, in a way that does not incur substantial damage in human cell membranes.
“Our substantially augmented deep learning approach allowed us to predict this new structural class of antibiotics and enabled the finding that it is not toxic against human cells.”