Hunting antibiotics with AI agents: Google DeepMind searches protein data for peptides
Google DeepMind shows how its agent software Antigravity searches public databases for antimicrobial peptides. A first candidate is being tested in the lab – nothing is confirmed yet.
Photo: HansN. / Wikimedia Commons, CC BY-SA 3.0 (cropped and resized)
Video: YouTube · Google Antigravity · youtube.com
Google DeepMind on Thursday published a walkthrough of how the agent software Google Antigravity can be used to search for new antimicrobial peptides – small proteins that can kill bacteria. The backdrop is growing antibiotic resistance, which according to an estimate cited in the post could cause around ten million deaths a year by 2050.
The AI agent combs public resources such as the protein database UniProt, which holds millions of sequences of unknown function, the AlphaFold structure database and the search tool Foldseek. In the example shown, it found a 97-amino-acid bacteriocin of the Lactococcin 972 family from the bacterium Streptococcus pseudopneumoniae; respiratory pathogens are named as possible targets.
DeepMind stresses that the candidate is still undergoing wet-lab testing and its activity is unconfirmed. An accompanying video on the official Antigravity channel shows the process step by step.