At MIT, a team of researchers is utilizing deep learning—a type of artificial intelligence—to discover new, potentially life-saving antibiotics. Their focus is on combating one of the world’s deadliest drug-resistant bacterium: methicillin-resistant Staphylococcus aureus (MRSA), which takes over 10,000 lives in America annually.
Published in Nature, MIT’s study reveals that a new class of compounds, identified using deep learning models, can kill lab-grown MRSA. These compounds also show a very low toxicity against human cells—making them promising candidates for new antibiotics.
The team gathered data by testing nearly 39,000 compounds against MRSA. The chemical structure and antibiotic activity data of these compounds were then used to train the deep learning model. Given a new molecule, the model can provide a probability of that compound being antibacterial.
A distinctive aspect of this study was the successful application of an algorithm known as Monte Carlo tree search. This computational tool was utilized to make the deep learning model’s predictions more transparent. The model can now offer an estimate for each molecule’s antimicrobial activity and predict the influence of the molecule’s substructures on that activity.
A subsequent screening of about 12 million compounds helped researchers identify classes of molecules effective against MRSA. They purchased around 280 of these compounds for testing, resulting in the discovery of two promising antibiotic candidates.
This study is part of the Antibiotics-AI Project at MIT, aiming to discover new classes of antibiotics against seven deadly bacteria types over seven years. As part of the project, the researchers shared their findings with Phare Bio, a non-profit organization. The next steps in this research include conducting a more detailed analysis of the newly discovered compounds’ chemical properties and potential clinical use.
The study’s authors anticipate that their approach—based on the understanding of chemical substructures—will be beneficial in designing new compounds and discovering new classes of antibiotics against different pathogens. Similar strategies are already being employed by them in the hopes of finding further drug candidates.
This ambitious and vital mission is funded by the Antibiotics-AI Project, backed by various institutions, foundations, and an anonymous donor. Other contributors to the study include Integrated Biosciences Inc., the Wyss Institute for Biologically Inspired Engineering, and the Leibniz Institute of Polymer Research in Dresden, Germany.
