Google Antigravity Accelerates Antimicrobial Peptide Discovery

Google has outlined a workflow in Google Antigravity designed to accelerate the discovery of antimicrobial peptides, or AMPs, at a time when rising antimicrobial resistance is expected to cause millions of deaths annually if no effective alternatives are found. The briefing centers on AMPs, especially bacteriocins, which are viewed as promising substitutes for conventional antibiotics because they target essential bacterial functions and may make resistance harder for pathogens to acquire. The core of the workflow is ProtNLM, or Protein Natural Language Model, a deep learning system developed by Google to predict biological function directly from raw amino acid sequences. ProtNLM is trained to interpret evolutionary patterns embedded in protein sequences, allowing it to label previously uncharacterized entries in large databases such as UniProt. This is significant because billions of natural amino acid sequences have been generated by genomic and metagenomic sequencing, but many remain in the unannotated “dark matter” of protein databases, with no clear functional description. By assigning high-confidence functional labels to these sequences, ProtNLM can help researchers surface promising AMP candidates at scale. Google also emphasized that database mining alone is not enough. To support end-to-end analysis, Antigravity now includes a Biomolecular Structure Viewer, a three-dimensional visualization tool embedded directly into the collaborative workspace. The viewer allows researchers and AI agents to work over a shared structural context. Agents can retrieve predicted structures from the AlphaFold Database or from a scientist’s own models, highlight functional regions, and use the 3D view to explain structural hypotheses. User interactions are synchronized back to the agent, meaning a researcher can pin notes, select residues, or mark a binding pocket in 3D space and ask the agent to reason about that exact structural context. The example presented in the briefing involves a candidate derived from Streptococcus pseudopneumoniae. The candidate is described as biologically plausible and supported by structural and genomic evidence, but Google stresses that it still requires wet-lab validation. The broader value of the demo lies in its workflow implications: instead of days of fragmented database searches, researchers can guide an AI agent through targeted, interactive biological reasoning, adjusting filters, sequence lengths, and pathogen priorities in real time. The approach may extend beyond human commensals to environmental microbes and other priority disease targets.