The urgent need for new antibiotics stems from the accelerating threat of antimicrobial resistance. Antimicrobial peptides (AMPs), especially bacteriocins, are attractive because they naturally target essential bacterial structures, making resistance development difficult. Massive genomic and metagenomic sequencing has filled public repositories such as UniProt with billions of raw amino‑acid sequences, many of which remain unannotated "dark matter" that could hide novel AMPs.
Google Antigravity offers ProtNLM (Protein Natural Language Model), a deep‑learning system that reads the evolutionary grammar of protein sequences and directly predicts functional annotations. This enriches millions of unreviewed UniProt entries with high‑confidence labels. In Antigravity, researchers set a focused goal: discover and computationally validate new AMPs in Streptococcus pseudopneumoniae.
During an interactive multi‑turn session, the AI agent performed the following core steps:
-
Candidate filtering – queried UniProt and narrowed down to a 97‑aa peptide from the Lactococcin 972 family (UniProt ID: A0A0U0D591).
-
Novelty check – used the EMBL‑EBI NCBI‑BLAST service to compare the sequence against curated Swiss‑Prot entries, confirming no meaningful homologs and establishing novelty.
-
3D structure reasoning – retrieved the candidate’s model from the AlphaFold Protein Structure Database, aligned it with experimentally solved structures, and observed a fold closely matching known bacteriocins, reinforcing ProtNLM’s functional prediction.
-
Genomic context analysis – examined neighboring genes and found the candidate positioned just upstream of an operon containing an ABC transporter permease, its ATP‑binding partner, another annotated bacteriocin, and an XRE transcriptional regulator, indicating a functional bacteriocin module.
-
Literature‑driven hypothesis – surveyed existing studies and proposed a mechanism targeting bacterial cell division, with predicted activity against several drug‑resistant pathogens.
-
Experimental plan generation – produced a concrete wet‑lab protocol covering gene cloning, protein expression, antimicrobial susceptibility testing, target engagement assays, and in‑vitro toxicity assessment; the plan is now being executed.
Antigravity also embeds a Biomolecular Structure Viewer, allowing researchers and the AI to share a 3D canvas. Users can highlight residues or pin notes directly on the structure, and the agent instantly incorporates these cues into its reasoning (e.g., “How might this β‑strand affect Lipid II binding?”). This bidirectional interaction eliminates the need for separate visualization tools.
By collapsing what traditionally required multiple specialized software packages, custom scripts, and manual data transfers into a single collaborative session, Google Antigravity dramatically lowers the barrier for routine biological workflows. Researchers can easily adjust filters, explore different peptide lengths, or target other pathogens, enabling scalable AMP discovery across diverse microbial niches.
Review: The integration of deep‑learning‑based function prediction, structural retrieval, and interactive visualization in Google Antigravity provides a seamless end‑to‑end pipeline for antimicrobial peptide discovery, accelerating the transition from in‑silico candidates to experimental validation.