Researchers have harnessed the power of more transparent deep learning models to uncover a novel class of new antibiotics targeting drug-resistant Staphylococcus aureus (MRSA) bacteria, marking a significant breakthrough in combating antibiotic resistance. The application of artificial intelligence (AI) has revolutionized medicine, enabling scientists to unveil the first new antibiotics in six decades.
The breakthrough revolves around a newly discovered compound capable of eradicating a drug-resistant bacterium responsible for thousands of fatalities globally each year, potentially reshaping the landscape in the fight against antibiotic resistance.
James Collins, a professor of Medical Engineering and Science at the Massachusetts Institute of Technology (MIT) and a co-author of the study, emphasized the innovation: “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.” Collins described their work as introducing a time-efficient, resource-efficient, and mechanistically insightful framework, shedding light on chemical structures in unprecedented ways.
When it comes to medicine with the technology now helping scientists to unlock the first new antibiotics in 60 years.
Published in Nature, the research involved a team of 21 scientists dedicated to unraveling the mysteries of antibiotic discovery. The project employed a deep-learning model to predict the activity and toxicity of the groundbreaking compound. Deep learning, utilizing artificial neural networks to autonomously learn and represent features from data, has gained prominence in drug discovery for its ability to accelerate identification, predict properties, and optimize the drug development process.
The focus of the study was on methicillin-resistant Staphylococcus aureus (MRSA), a formidable pathogen causing conditions ranging from mild skin infections to life-threatening pneumonia and bloodstream infections. The European Centre for Disease Prevention and Control (ECDC) reports nearly 150,000 annual MRSA infections in the European Union, leading to approximately 35,000 deaths from antimicrobial-resistant infections.
The MIT research team expanded on the conventional deep learning model by training an extensively enlarged version using expanded datasets. Evaluating approximately 39,000 compounds for their antibiotic activity against MRSA, they input both the resulting data and the chemical structures of the compounds into the model.
Felix Wong, a postdoc at MIT and Harvard and one of the lead authors, explained the objective of the study: “What we set out to do in this study was to open the black box.” The researchers sought to demystify the intricate calculations underlying these models, which emulate neural connections and remain largely opaque.
To refine the selection of potential drugs, the researchers integrated three additional deep-learning models designed to assess compound toxicity on three distinct types of human cells. Combining toxicity predictions with antimicrobial activity data, the team identified compounds capable of effectively combating microbes with minimal harm to the human body. This set of models screened approximately 12 million commercially available compounds and pinpointed promising candidates from five different classes, categorized based on specific chemical substructures within the molecules, that exhibited predicted activity against MRSA.