The Google Antigravity platform unified millions of proteins in the UniProt database to rapidly identify new antibiotics. The ProtNLM model predicted the peptide's activity against respiratory tract pathogens.

#AntimicrobialResistance #GoogleAntigravity #Protnlm

구글 딥마인드, 항균 펩타이드 탐색 사례 공개

구글 딥마인드가 Google Antigravity의 AI 에이전트와 연구자가 한 세션에서 새 항균 펩타이드 후보를 찾아낸 사례를 공개했다. UniProt의 기능 미상 단백질을 ProtNLM으로 훑어 고른 박테리오신 후보는 지금 습식 실험 검증을 받고 있다.

How to look under the hood of protein sequence functional annotations? We have a DUF that is sometimes labeled with 2 different types of "___ase". One traces back to google's ProtNLM (natural language model) (I think), the other ... ? No clear structural homologs with experimental functions.

Any specific taxa? Uniprot relies on Ensembl/Refseq genome annotations. (At some point, they also incorporated ProtNLM for their annotation pipeline) If it is not some exotic species without any reference proteome, you may find Heng Li's miniprot good enough to do the job.