Overview
- Two Science papers published Thursday used machine learning to scan tens of thousands of bacterial genomes and flagged a huge set of antiphage proteins.
- A Pasteur-led deep-learning sweep predicted 2.39 million antiviral proteins across more than 30,000 genomes, with about 85% not previously tied to immunity.
- MIT researchers built DefensePredictor, which in 69 E. coli strains identified 624 candidate defense proteins and confirmed 42 of 94 tested genes blocked at least one bacteriophage virus.
- The deep-learning analysis estimated that roughly 1.5% of genes in a typical bacterium defend against phages, showing that past counts missed much of this immune space.
- The teams released the DefensePredictor software and an interactive antiphage atlas, enabling labs to validate untested systems that could one day yield tools beyond CRISPR and restriction enzymes.