Google says its Antigravity agentic development platform can bring several stages of antimicrobial peptide research into one workspace, helping biologists search databases, assess protein structures and rank potential candidates. In a technical walkthrough, researchers used the platform to identify a peptide candidate in Streptococcus pseudopneumoniae with predicted activity against pathogens involved in respiratory tract infections. The finding has not yet been confirmed in laboratory tests and is undergoing wet-lab validation.
The search draws on ProtNLM, Google’s protein language model, which predicts biological function from amino acid sequences. The model is intended to help researchers examine previously uncharacterized entries in UniProt, a public database containing millions of protein sequences with unknown functions.
Google presents antimicrobial peptides, including bacteriocins, as possible alternatives to conventional antibiotics. These naturally occurring molecules can target vulnerable features of competing bacteria. The company says the growing volume of genomic and metagenomic data offers a large pool of sequences to investigate, although many remain unannotated because they do not closely resemble known proteins.
In the demonstrated workflow, a researcher set the goal of finding and computationally validating new antimicrobial peptides in S. pseudopneumoniae. The researcher guided the investigation, while Antigravity coordinated database searches, structural analysis and literature review. Google says the candidate is supported by genomic and structural evidence, but stresses that experimental validation is still needed.
The platform also includes a Biomolecular Structure Viewer, which displays protein structures in 3D. Researchers and the AI agent can work with the same visual workspace: the agent can highlight regions to explain its hypotheses, while scientists can mark residues or add notes that inform the agent’s analysis.
Google says the integrated approach is designed to reduce the time spent moving between databases, scripts and modelling tools. The company describes the walkthrough as a way to narrow the search to promising leads before resource-intensive laboratory testing, and says researchers can adapt the workflow to different sequence criteria or target organisms.
Sources: Google DeepMind Blog.