CAMBRIDGE, MA - In a groundbreaking study, MIT researchers have utilized deep learning, a type of artificial intelligence, to identify a class of compounds with the potential to combat methicillin-resistant Staphylococcus aureus (MRSA), a drug-resistant bacterium responsible for over 10,000 deaths annually in the United States.

Published today in Nature, the research demonstrates that these compounds exhibit the ability to eradicate MRSA in laboratory settings and in two mouse models infected with MRSA. Importantly, these compounds also boast low toxicity towards human cells, positioning them as strong candidates for drug development.

An innovative aspect of the study lies in the researchers' ability to understand the underlying factors driving the deep learning model's antibiotic potency predictions. This newfound knowledge could pave the way for the design of more effective drugs targeting MRSA.

Led by James Collins, the Termeer Professor of Medical Engineering and Science at MIT's Institute for Medical Engineering and Science and Department of Biological Engineering, the study's lead authors are Felix Wong, postdoc at MIT's Institute for Medical Engineering and Science and the Broad Institute of MIT and Harvard, and Erica Zheng, formerly a graduate student at Harvard Medical School.

MRSA is responsible for infecting over 80,000 individuals annually in the United States, typically causing skin infections and pneumonia. Severe cases can lead to sepsis, a life-threatening bloodstream infection.

Collins and his team have been utilizing deep learning in recent years to explore new avenues for antibiotic development. Prior work by the researchers has yielded potential treatments for drug-resistant bacteria such as Acinetobacter baumannii, commonly found in hospital settings.

The compounds identified in this study were discovered through deep learning models capable of recognizing chemical structures associated with antimicrobial activity. By sifting through millions of compounds, these models generated predictions regarding their antimicrobial potential.

While these searches have proven fruitful, one limitation has been the lack of transparency in understanding the models' decision-making processes. Opening this "black box" would greatly enhance the identification and design of future antibiotics.

In order to reveal the inner workings of their model, the researchers adapted an algorithm called Monte Carlo tree search. This algorithm provides insights not only into the estimated antimicrobial functions of each molecule but also predicts the specific substructures within the molecule that contribute to its activity.

To further narrow down the selection of potential drugs, three additional deep learning models were trained to assess the compounds' toxicity towards different types of human cells. By combining this information with the antimicrobial activity predictions, the researchers successfully identified compounds capable of selectively killing microbes while minimizing harm to human cells.

Employing these models, the researchers screened approximately 12 million commercially available compounds. From this vast collection, they pinpointed compounds from five distinct classes, each featuring chemical substructures predicted to be effective against MRSA.

Out of the 280 compounds purchased for further testing, two compounds from the same class exhibited significant promise as antibiotic candidates. In mouse models infected with MRSA, these compounds reduced MRSA populations tenfold.

Experiments conducted by the researchers demonstrated that these compounds target bacterial cell membranes, disrupting their ability to maintain an electrochemical gradient crucial for essential cellular functions. Notably, these compounds do not inflict substantial damage on human cell membranes, suggesting a selective mechanism of action against bacterial cells.

The research findings have been shared with Phare Bio, a nonprofit organization established by Collins and others as part of the Antibiotics-AI Project. The nonprofit intends to conduct a comprehensive analysis of the compounds' chemical properties and their potential clinical applications. Meanwhile, Collins' lab will continue designing additional drug candidates based on the study's findings, while also exploring compounds with the potential to combat other types of bacteria.

The research received funding from various sources, including the James S. McDonnell Foundation, the U.S. National Institute of Allergy and Infectious Diseases, the Swiss National Science Foundation, and the U.S. National Institutes of Health, among others. The Antibiotics-AI Project is supported by the Audacious Project, Flu Lab, the Sea Grape Foundation, the Wyss Foundation, and an anonymous donor.

This groundbreaking study represents a significant step forward in the fight against drug-resistant bacteria. Through the synergy of artificial intelligence and molecular research, these MIT researchers have brought us closer to finding effective treatments for MRSA and potentially other deadly bacterial infections.