|
Getting your Trinity Audio player ready...
|
MIT researchers have leveraged the power of artificial intelligence, specifically deep learning, to uncover a class of compounds with the potential to combat drug-resistant bacteria. In their study, the researchers demonstrated the ability of these compounds to effectively eliminate methicillin-resistant Staphylococcus aureus (MRSA), a bacterium responsible for over 10,000 deaths annually in the United States.
MRSA, prevalent in more than 80,000 cases in the U.S. each year, often leads to skin infections, pneumonia, and, in severe patients, sepsis—a life-threatening bloodstream infection. Addressing the urgent need for new antibiotics, MIT’s Antibiotics-AI Project, led by Professor James Collins, aims to discover novel antibiotic classes against seven types of deadly bacteria within a seven-year timeframe.
The researchers focused on MRSA, employing deep learning models to identify chemical structures associated with antimicrobial activity. While successfully generating potential drugs against various drug-resistant bacteria, the challenge with these models lies in their “black box” nature—making it unclear on what basis the models make predictions.
In this study, the MIT team sought to unveil the inner workings of the deep-learning model. The first step involved training an expanded dataset of approximately 39,000 compounds tested for antibiotic activity against MRSA. This data, combined with information on the chemical structures of the compounds, was fed into the model, which was trained to recognise antibacterial properties.
To elucidate the model’s prediction rationale, the researchers employed the Monte Carlo tree search algorithm. This algorithm estimated each molecule’s antimicrobial activity and predicted the substructures likely responsible for that activity.
The next crucial step was narrowing down the pool of candidate drugs. Three additional deep-learning models were trained to predict the toxicity of the compounds to different types of human cells. Integrating this toxicity information with predictions of antimicrobial activity enabled the identification of compounds capable of killing microbes with minimal adverse effects on human cells.
Screening about 12 million commercially available compounds, the models pinpointed compounds from five classes with predicted activity against MRSA based on chemical substructures. Of the 280 compounds purchased and tested against MRSA in a lab dish, two from the same class emerged as up-and-coming antibiotic candidates.
In mouse models—one simulating MRSA skin infection and another simulating MRSA systemic infection—each identified compound reduced the MRSA population by 10. Importantly, experiments revealed that these compounds disrupted the bacteria’s ability to maintain an electrochemical gradient across their cell membranes, a vital factor for critical cell functions.
The study suggests a promising avenue for combating MRSA, showcasing strong evidence that the newly discovered structural class selectively targets Gram-positive pathogens by dissipating the proton motive force in bacteria. The compounds selectively attack bacterial cell membranes without causing substantial damage to human cell membranes.
The researchers have shared their findings as part of the Antibiotics-AI Project. They plan to conduct a more detailed analysis of these compounds’ chemical properties and potential clinical use. Meanwhile, Collins’s lab is actively designing additional drug candidates based on the study’s findings and utilising the models to identify compounds effective against other types of bacteria.
Felix Wong, a postdoc at MIT’s Institute for Medical Engineering and Science (IMES) and the Broad Institute of MIT and international university, and Erica Zheng, a Medical School graduate student from the international university advised by Collins, served as the study’s lead authors. Their work represents a significant advancement in antibiotic discovery, marrying deep learning capabilities with a nuanced understanding of the chemical structures that drive antimicrobial activity.
Professor James Collins emphasised the novel insights from the study, stating, “The insight here was that we could see what was being learned by the models to make their predictions that certain molecules would make for good antibiotics. This 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.” Wong added that the study aimed to “open the black box” of deep learning models, revealing the intricate calculations mimicking neural connections.
MIT’s research marks a significant step in antibiotic discovery, offering a more transparent and insightful approach to combatting antibiotic-resistant pathogens. The collaborative efforts of the Antibiotics-AI Project highlight the transformative impact of digital technology in the realm of healthcare and medical research, setting the stage for continued advancements in the fight against infectious diseases.
