Accelerating the search for antimicrobial peptides with Google Antigravity

Accelerating Antimicrobial Peptide Discovery with Google Antigravity

Antimicrobial resistance threatens global health, and the hunt for new therapeutics—especially antimicrobial peptides (AMPs)—has traditionally been slow and fragmented. Google Antigravity, an agentic development platform, reshapes this workflow by stitching together database mining, deep‑learning annotation, structural modeling, and experimental design into one collaborative environment.

A Unified Search Across Millions of Sequences

The core of Antigravity’s power is ProtNLM, a protein‑language model trained to read the evolutionary “grammar” of amino‑acid strings. ProtNLM can assign high‑confidence functional labels—names, GO terms, subcellular locations—to the vast swathes of UniProtKB/TrEMBL entries that lack manual annotation. In practice, a researcher can ask the AI to “find novel AMPs in Streptococcus pseudopneumoniae” and receive a filtered list of candidates, each accompanied by predicted activity and confidence scores.

Collaborative Intelligence in Action

The platform encourages a multi‑turn dialogue between scientist and agent:

1. Goal definition – The user sets the biological objective (e.g., locate bacteriocins).
2. Database orchestration – The AI queries UniProt, applies ProtNLM predictions, and narrows millions of sequences to a handful of high‑potential hits.
3. Novelty verification – Using integrated BLAST services, the agent confirms that a candidate (e.g., UniProt A0A0U0D591) lacks close homologs in curated Swiss‑Prot, indicating true novelty.
4. Structural reasoning – The AI pulls the candidate’s AlphaFold model, aligns it with experimentally solved structures, and highlights fold similarity to known bacteriocins, providing structural corroboration of the functional prediction.
5. Genomic context – By inspecting neighboring genes, the agent identifies operon features (ABC transporter, XRE regulator) that support a bacteriocin role.
6. Literature synthesis – The AI scans recent papers, proposes a mechanism (e.g., inhibition of cell division), and suggests a pathogen spectrum.
7. Experimental blueprint – Finally, Antigravity generates a step‑by‑step wet‑lab plan covering expression, susceptibility testing, and toxicity assays.

This end‑to‑end session compresses what would normally require days of manual triage across multiple tools into a single, coherent conversation.

The Biomolecular Structure Viewer: A Shared 3D Canvas

Understanding a peptide’s fold is essential for hypothesis generation. Antigravity embeds a Biomolecular Structure Viewer that synchronizes with the AI’s actions. When the agent retrieves an AlphaFold model, it can automatically highlight functional motifs, overlay homologous structures, and let the researcher manipulate the view in real time. This bidirectional interaction ensures that visual insights directly inform the AI’s reasoning and vice versa.

Scaling Beyond a Single Species

While the walkthrough focused on S. pseudopneumoniae, the workflow is agnostic to organism or peptide length. Researchers can iteratively adjust filters—such as charge, hydrophobicity, or secretion signals—to explore environmental metagenomes, human microbiomes, or synthetic libraries. By leveraging ProtNLM’s ability to annotate the “dark matter” of protein space, Antigravity opens a scalable path to discover AMPs across the entire UniProt universe.

Practical Takeaways

• Speed: From query to experimental plan in under an hour, versus weeks of manual curation.
• Confidence: Integrated functional, structural, and genomic evidence reduces false leads.
• Accessibility: No need for separate bioinformatics pipelines; the platform handles cross‑database calls, model retrieval, and literature mining.
• Collaboration: The shared 3D viewer and conversational interface keep human intuition at the forefront while offloading repetitive computation to the AI.

Google Antigravity demonstrates how agentic AI can democratize complex bio‑discovery workflows, turning the vast, unannotated protein landscape into a fertile hunting ground for the next generation of antimicrobial peptides.