Google’s Antigravity Workflow Finds Antimicrobial Peptide Candidate

Google’s Antigravity platform linked scientific databases with DeepMind’s ProtNLM model to surface an antimicrobial peptide candidate, illustrating how AI agents can connect steps in drug research.

In an Oct. 15 technical demonstration, researchers used ProtNLM to search millions of unannotated UniProt entries. UniProt is a public protein database, and many of the searched entries had no known function attached. Antimicrobial peptides are short chains of amino acids that can fight bacteria.

The candidate was flagged for promising activity against Streptococcus pseudopneumoniae, a respiratory pathogen. Researchers used AlphaFold to predict its protein structure on a computer, a process known as an in silico check. The work produced a plan for wet-lab experiments, but physical tests have not confirmed the peptide’s activity.

Antigravity brings together more than 30 scientific databases and Google DeepMind models. Its agent layer can connect research steps and produce outputs such as 3D renderings and experimental plans without requiring scientists to write code. The platform is intended to shorten complex research workflows that can take days into minutes.

The workflow builds on Google’s Science Skills suite, introduced in May 2026, which includes more than 11 tools for proteomics and cheminformatics. This demonstration focused on peptide searching and AlphaFold structure prediction. TokenPost previously covered Antigravity in a music-synchronized 3D periodic-table project.