Google's AI Agent Mines Antibiotic Leads in One Session
Google Antigravity compresses days of fragmented database work into a single AI-guided session—and has already flagged a wet-lab candidate.
By The ai-compute Desk
Antimicrobial resistance could kill an estimated 10 million people annually by 2050 [1][2]. Google thinks an agentic AI platform can help close the discovery gap before that clock runs out.
Google Antigravity provides an agentic science workbench that compresses what used to be a fragmented search over multiple databases into a single session [1]. In a published walkthrough, researchers used it to identify a bacteriocin candidate—97 amino acids from the Lactococcin 972 family—found among the millions of sequences with unknown functionality held in UniProt [1].
The platform's ProtNLM model predicts biological function directly from raw protein sequences, assigning high-confidence functional labels to millions of previously unnamed entries [1]. In this case, structural analysis via the AlphaFold Protein Structure Database confirmed the candidate's fold closely matches an experimentally determined bacteriocin, while genomic context—its position just 6 base pairs ahead of an ABC transporter permease—supported the hypothesis that it belongs to a working bacteriocin system [1].
The agent then generated a bench protocol spanning protein expression, antimicrobial susceptibility testing, target engagement, and in vitro toxicity testing [1]. That candidate, along with several others discovered through the same pipeline, is now in wet-lab validation [1].
The real pressure here is temporal. Drug discovery has always been slow; the biology of resistance is not.
This session focused on a narrow amino acid window in Streptococcus pseudopneumoniae, but the architecture is built to range wider—across human commensals, environmental microbes, and beyond [1]. Whether speed of hypothesis generation translates to speed of clinical impact is the test wet labs will now have to answer.
turning days of fragmented database triage into a single collaborative sessionPengfei Tian et al.