Researchers from MIT have identified a new class of antibiotics that could potentially target Methicillin-resistant Staphylococcus aureus (MRSA), a drug-resistant bacterium accountable for over 10,000 deaths in the US annually. A significant contribution of the study is the understanding of the deep learning model that predicts antibiotic potency. The insights gained could aid in designing drugs that are even more effective.
The researchers applied deep learning models trained to recognize chemical structures associated with antimicrobial effects to millions of other compounds, predicting those likely to demonstrate strong antimicrobial activity. These ‘black box’ models have limitations, primarily because their prediction mechanisms remain unknown to scientists. Understanding this process could facilitate the identification or design of additional antibiotics.
To decode model predictions, researchers substantially expanded their dataset. They tested approximately 39,000 compounds for antibiotic activity against MRSA. They fed this data, along with information on the compounds’ chemical structures, into the model. An algorithm known as the Monte Carlo tree search was applied to make the working of the model more explainable. This allowed the model to estimate each molecule’s antimicrobial activity and predict which substructures of the molecules likely accounted for that activity.
Alongside antimicrobial activity, the researchers trained additional models to predict compounds’ toxicity to three types of human cells. From an initial screen of 12 million compounds, they identified two promising antibiotic candidates from one chemical class.
Tests showed that the compounds disrupt bacteria’s ability to maintain an electrochemical gradient across their cell membranes, which is critical for key cell functions. This mechanism is similar to an antibiotic identified by the same lab in 2020 but targets Gram-positive bacteria, like MRSA, that have thicker cell walls.
The researchers have shared their findings with Phare Bio, a nonprofit part of the MIT Antibiotics-AI Project. The nonprofit plans further analysis of the compounds’ chemical properties and potential clinical use. Meanwhile, Collins’ lab is working on additional drug candidates based on study findings and using the models to seek compounds effective against other types of bacteria.
