Using artificial intelligence (AI) technology called deep learning, MIT researchers have identified compounds capable of defeating methicillin-resistant Staphylococcus aureus (MRSA), a drug-resistant bacterium causing over 10,000 deaths annually in the US. The compounds, which exhibit low toxicity to human cells, were found to effectively kill MRSA in lab and mouse models, making them potential drug candidates.
The key innovation of the study, published in Nature, is the researchers’ ability to understand the information the AI used to predict antibiotic efficacy. This knowledge could aid in designing improved drugs. This groundbreaking work forms part of the Antibiotics-AI Project at MIT, led by James Collins, which aims to discover new classes of antibiotics against seven deadly bacteria types over seven years.
Deep learning models were used to learn how to identify chemical structures associated with antimicrobial activity, then search through millions of other compounds to predict those with the highest antimicrobial potential. Pulling back the curtain on this process was a significant achievement of this study, allowing scientists to see under the hood of the black-box deep learning models and better understand their predictions.
The research team trained the deep learning model with vast datasets, testing about 39,000 compounds for antibiotic activity against MRSA. They also adapted an algorithm called Monte Carlo tree search, which made the model’s predictions more explainable. This allowed the model to predict not only a molecule’s antimicrobial activity but also what parts of the molecule likely contribute to this activity.
Finally, the researchers used the model to sift through about 12 million commercially available compounds, finding five classes of compounds predicted to be active against MRSA. Further testing revealed two very promising antibiotic candidates that showed potent activity against MRSA in mouse models and appeared to work by disrupting the bacteria’s ability to maintain an electrochemical gradient across their cell membranes.
This discovery is a significant advancement in AI’s role in medicine and antibiotic development. It also highlights the immense potential of AI and machine learning in addressing complex medical challenges. Collins’ team is currently working on designing new drug candidates based on the study’s findings and using the models to uncover compounds that can kill other types of bacteria. Furthermore, the findings have been shared with Phare Bio, a non-profit, also part of the Antibiotics-AI Project, which plans to examine the chemical properties and potential applications of these compounds in greater detail.
