UMass Amherst Pairs AI Neural Network with High-Throughput Screening to Accelerate Tuberculosis Drug Discovery

Researchers at the University of Massachusetts Amherst, led by Assistant Professor Anna Green, published a paired-technique approach on July 6, 2026 to accelerate the search for tuberculosis drugs, in Nature Microbiology. The team combined PAC-MAN (Peptidoglycan Accessibility Click-Mediated AssessmeNt), a high-throughput laboratory screening method, with MycoPermeNet, a machine-learning model that predicts which chemical compounds can penetrate the notoriously impermeable outer membrane of Mycobacterium tuberculosis. The AI model was trained on PAC-MAN's experimental measurements of which compounds cross the bacterium's protective mycomembrane, then used to predict permeability for compounds not directly tested in the lab — dramatically narrowing the search space for new TB antibiotics. Tuberculosis remains the world's deadliest single-agent infectious disease, killing an estimated 1.23 million people in 2024; its resistant outer membrane has long blocked many otherwise-promising antibiotic candidates from working.