Researchers from MIT have employed deep learning artificial intelligence (AI) to discover a set of compounds capable of exterminating methicillin-resistant Staphylococcus aureus (MRSA), a drug-resistant bacterium causing over 10,000 deaths annually in the U.S. Published in Nature, the study highlights that these compounds can kill MRSA both in a lab and in two MRSA mouse infection models. The low toxicity of these compounds against human cells makes them suitable drug candidates.
Notably, this study enabled researchers to understand the information utilized by the deep-learning model to predict antibiotic efficacy. This knowledge can guide researchers in crafting more effective drugs. This method offers a time-efficient and resource-efficient approach, providing insights from a chemical-structure standpoint, previously unavailable.
Driving this research is MIT’s Antibiotics-AI Project, led by James Collins, aiming to unearth new antibiotic classes against seven lethal bacteria types within seven years. However, current AI models are often “black boxes,” providing no details on the features the model used for prediction. By uncovering these features, researchers could efficiently discover and design additional antibiotics.
Wong and his colleagues trained the deep-learning model using large datasets, containing information about each compound’s chemical structures and its antibiotic capacity against MRSA. The model can then predict a new compound’s probability of being antibacterial. To make these predictions more transparent, the team used a Monte Carlo tree search algorithm to generate both an estimate of each molecule’s antimicrobial activity and which substructures likely contribute to that activity.
Using these models, the researchers screened around 12 million commercially available compounds and found five different compound classes likely to be active against MRSA. Two of these five displayed promising antibiotic potential, reducing the MRSA population by a factor of ten in MRSA skin infection and systemic infection mouse models. The compounds operate by disrupting the bacteria’s capability to upkeep its electrochemical gradient across cell membranes, crucial for ATP production—a process that stores energy in cells.
Scientists suggest that these newly discovered antibiotics target bacterial cell membranes in a way that does not induce substantial harm in human cell membranes. Further, this development was made possible due to their enhanced deep learning technique. Collins’ team is now concentrating on designing more drug candidates based on this study’s discoveries, whilst using the models to find compounds capable of eliminating other bacteria types.
The findings have been shared with Phare Bio, a nonprofit started by Collins and others as part of the Antibiotics-AI Project. Plans are underway to conduct a more detailed analysis of the chemical properties and potential clinical use of these compounds.
