🧬Google Antigravity Mines Millions of Unknown Proteins for New Antibiotics

Your antibiotic pipeline just got a computational shortcut

Google just pointed its AI at one of the nastiest problems in medicine, and the first candidate is already on a lab bench. Antigravity, Google's agentic science workbench, let a researcher and an AI agent hunt for novel antimicrobial peptides in a single session, from raw sequence to bench protocol. If you work in computational biology, this is a workflow change. Normally finding a viable protein means bouncing between databases, structure tools, and literature, then rebuilding the context each time. Here the agent handles the search, structural comparison against AlphaFold and experimental structures, and genomic neighborhood reasoning, all in the same place. The Biomolecular Structure Viewer lets the human and agent steer the same 3D model together. The numbers matter. ProtNLM reads amino acid sequences and predicts function, enriching millions of previously unnamed UniProtKB/TrEMBL entries with functional labels covering protein names, GO terms, and subcellular locations. The agent's first pick was a 97-amino-acid bacteriocin from the Lactococcin 972 family, ID A0A0U0D591, targeting Streptococcus pseudopneumoniae. Several other candidates from the same pipeline are also heading into wet-lab testing. What to watch: predictions are not validation. The real test is whether expression, susceptibility, and toxicity assays confirm what the model suggests.