Google Antigravity Finds Novel 97 Amino Acid Antimicrobial Peptide
Google Antigravity found a 97 amino acid bacteriocin candidate, using ProtNLM, AlphaFold and genomic evidence before wet lab validation begins in 2026.
Summary
On October 15, 2026, Pengfei Tian, Maxwell L. Bileschi, Zoi Tsangalidou, Sukhdeep Singh and Lucy J. Colwell detailed Google Antigravity, an agentic science workbench combining ProtNLM with UniProt, structural databases and literature tools. ProtNLM reads raw amino acid sequences to assign protein names, function comments, subcellular locations, GO terms and keywords to millions of unreviewed UniProtKB/TrEMBL entries. Antimicrobial resistance could cause an estimated 10 million deaths annually by 2050, while billions of natural sequences include established or genome mined bacteriocins such as Nisin, Darobactin, Dynobactin and Lariocidin.
In an interactive search targeting Streptococcus pseudopneumoniae, the researcher steered questions while the agent narrowed millions of proteins to a handful. Lead A0A0U0D591 is a 97 amino acid Lactococcin 972 family bacteriocin with no meaningful sequence homolog among manually reviewed Swiss-Prot proteins in an EMBL-EBI NCBI-BLAST search. Its AlphaFold Protein Structure Database model closely matched an experimentally determined bacteriocin fold. It sits 6 base pairs upstream of an ABC transporter permease and possible ATP binding partner, beside another bacteriocin and an XRE regulator, supporting a functional system. Literature suggests a cell division target and predicted activity against several drug resistant pathogens involved in respiratory tract infections.
Antigravity generated a bench plan covering protein expression, antimicrobial susceptibility, target engagement and in vitro toxicity. This lead and several others are undergoing wet lab validation, which remains decisive. Google also introduced the Biomolecular Structure Viewer, a synchronized three dimensional canvas where scientists and the agent can share AlphaFold or custom models, residue highlights, notes and binding pocket context. The workflow compresses days of custom scripts and disconnected database, genomic and modelling work into one session, and can expand to other sequence lengths, priority pathogens, human commensals and environmental microbes.
Positives
- A0A0U0D591 has structural and genomic evidence supporting its classification as a 97 amino acid Lactococcin 972 family bacteriocin.
- ProtNLM can add high confidence functional labels to millions of previously unnamed UniProt entries directly from raw amino acid sequences.
- Antigravity reduced millions of protein entries to a handful of candidates during one researcher guided session.
- The Biomolecular Structure Viewer synchronizes models, residue selections, notes and binding pocket context between scientists and the AI agent.
- Several candidates have progressed to wet lab testing for expression, antimicrobial activity, target engagement and toxicity.
Risks & concerns
- A0A0U0D591 remains a computationally predicted lead, and wet lab experiments have not yet confirmed its antimicrobial activity or safety.
- Antimicrobial resistance could cause an estimated 10 million deaths annually worldwide by 2050.
- Millions of UniProt sequences still lack characterized functions because they have little similarity to known proteins.
- Wet lab validation remains resource intensive despite Antigravity narrowing the search to higher confidence candidates.
- The proposed cell division mechanism and activity against drug resistant respiratory pathogens remain predictions derived from literature and computational evidence.