Deep Learning Model Screens Millions of Compounds, Discovers 6 New Antibiotics Effective Against Resistant Bacteria
-
A deep learning model was used to screen over 12 million compounds and identify promising antibiotics. The model was interpretable, allowing rationales for activity predictions to be extracted.
-
283 molecules were tested from model predictions, yielding 6 novel antibiotic scaffolds selective for Gram-positive pathogens like S. aureus.
-
2 scaffolds were found to alter membrane energetics. They cleared infections in mouse models, including resistant strains, without toxicity.
-
Whole genome sequencing showed that bacterial resistance readily arose to the new antibiotics, motivating combination therapies.
-
This work demonstrates that deep learning can accelerate antibiotic discovery, but continued innovation is still needed to fully address drug resistance.