Accelerating the search for antimicrobial peptides with Google Antigravity

Google DeepMind has published a technical walkthrough, dated October 15, 2026, showing how biologists used Google Antigravity, the company's agentic development platform, to search for new antimicrobial peptides. The authors, including Pengfei Tian and Lucy J. Colwell, pitch the tool as a way to fold a scattered, multi-database hunt into one session. It draws on ProtNLM, a Google model that predicts protein function from raw amino acid sequences, to mine millions of uncharacterized entries in UniProt.

In the example, a researcher steered an agent toward novel peptides in Streptococcus pseudopneumoniae. The agent surfaced a 97-amino-acid bacteriocin from the Lactococcin 972 family, and a BLAST check against Swiss-Prot found no meaningful homolog among manually reviewed proteins. An AlphaFold model closely matched an experimentally determined bacteriocin's fold, and neighboring genes, including an ABC transporter just 6 base pairs downstream, suggested a working bacteriocin system. From the literature, the agent proposed that the molecule targets bacterial cell division and may act on several drug-resistant pathogens, then drafted a lab protocol.

DeepMind stresses that the candidate's activity remains only predicted: it and several others from the same pipeline are in wet-lab validation, with no experimental results reported yet. The company also introduced a Biomolecular Structure Viewer inside Antigravity, where researcher and agent share annotations on 3D models.

The stakes, DeepMind notes, are rising antimicrobial resistance, which could cause an estimated 10 million deaths a year by 2050. Its claimed benefit is turning days of database triage into a single session, narrowing costly lab work to the strongest leads.