AI maps how a new antibiotic targets gut bacteria (news.mit.edu)

MIT CSAIL and McMaster researchers discovered a new narrow-spectrum antibiotic candidate, enterololin, that suppresses Escherichia coli strains linked to Crohn’s-like inflammation while sparing most of the gut microbiome. In mouse models the drug sped recovery and preserved microbial diversity compared with vancomycin. Crucially, the team used a generative AI docking model, DiffDock, to predict that enterololin binds the LolCDE lipoprotein-transport complex — a mechanistic lead that guided experiments rather than replaced them. The AI prediction was validated experimentally: lab-evolved enterololin-resistant E. coli had mutations in lolCDE, RNA-sequencing showed disrupted lipoprotein-transport pathways after treatment, and CRISPR knockdowns supported the same target. By combining high-throughput screening with DiffDock’s diffusion-based probabilistic docking, the researchers cut mechanism-of-action elucidation from years to roughly six months. They’ve released sequencing data and DiffDock-L code publicly, and Stokes’ spinout Stoked Bio is optimizing the compound for clinical development. Beyond a single molecule, this work demonstrates a shift in AI’s role in drug discovery — from finding hits to providing mechanistic hypotheses that accelerate validation. Rapid, reliable MOA mapping could lower costs and timelines for developing precision antibiotics, reduce collateral microbiome damage, and help combat antimicrobial resistance.