MIT researchers have leveraged the power of deep learning, a branch of artificial intelligence (AI), to discover a class of compounds that can potentially kill methicillin-resistant Staphylococcus aureus (MRSA). The discovery, described in a paper published in the journal Nature, saw the use of AI to predict the antibiotic potency of various molecules, an insight that could be instrumental in the development of novel drugs in the future.
As part of MIT’s Antibiotics-AI Project, led by James Collins, the researchers aim to uncover new antibiotics targeting seven lethal bacteria strains over a span of seven years. Collins, who co-authored the study, said: “Our work provides a framework that is time-efficient, resource-efficient, and mechanistically insightful, from a chemical-structure standpoint, in ways that we haven’t had to date”.
The team’s models utilized deep learning, applied to large datasets of compounds, to identify those associated with antimicrobial activity. However, one of the major drawbacks with the conventional use of AI, was the inability to identify the types of data the model would consider to make its predictions. Therefore, the research team adopted the Monte Carlo tree search algorithm, used to train other AI models such as AlphaGo, to discern what features the model based its predictions on. This method helped the research team understand which substructures of a compound contribute to its antimicrobial activity.
Following this, the researchers screened about twelve million compounds to identify candidates demonstrating minimal toxic effect against human cells while effectively combating harmful microbes. The model was able to pinpoint compounds from five different categories that would be effective against MRSA. Two promising antibiotics were identified and had significantly reduced the MRSA population in both a skin infection and systemic infection mouse model by a factor of ten.
The compounds seemed to work by compromising the bacteria’s ability to maintain an electrochemical gradient over their cell membranes, necessary for critical cell functions. Collins earlier discovered an antibiotic candidate, halicin, that works similarly but is specific to Gram-negative bacterial infections. The researchers aim to continue designing drugs based on these novel findings.
The results have been relayed to nonprofit firm Phare Bio for further analysis and potential clinical use of the compounds. Meanwhile, Collins’ laboratory continues to design drug candidates based on this research, leveraging similar methodologies to discover new classes of antibiotics against different pathogens.
Funding for the study was provided by the James S. McDonnell Foundation, the U.S. National Institute of Allergy and Infectious Diseases, the Banting Fellowships Program, the Volkswagen Foundation, the Defense Threat Reduction Agency, the U.S. National Institutes of Health, and the Broad Institute. Collins’ Antibiotics-AI Project is funded by various organizations including Flu Lab, the Wyss Foundation, and anonymous donors.
