Google DeepMind Introduces Antigravity AI Workbench for Peptide Discovery

What is new

Google DeepMind has introduced Antigravity, an agentic science workbench designed to assist biologists in the discovery and evaluation of antimicrobial peptides (AMPs) [1]. The platform uses a deep learning model called ProtNLM (Protein Natural Language Model) to predict biological functions from raw amino acid sequences [1]. This model is trained to read the evolutionary grammar of sequences, allowing it to provide functional labels for millions of uncharacterized protein entries in the UniProt database [1]. In a technical walkthrough, researchers used the platform to identify a novel bacteriocin candidate (Uniprot ID: A0A0U0D591) from the Lactococcin 972 family, which is 97 amino acids in length [1]. This specific lead is currently undergoing wet-lab validation [1].

How to use it

According to Google DeepMind, the discovery process involves an interactive, multi-turn session between a researcher and an AI agent [1]. The steps described include:

• Setting a focused biological goal, such as finding novel AMPs in Streptococcus pseudopneumoniae [1].
• The agent performs cross-database orchestration, structural analysis, and literature synthesis [1].
• The agent filters millions of database entries down to specific candidates [1].
• Researchers guide the agent to verify novelty by interrogating candidates against Swiss-Prot via the EMBL-EBI NCBI-BLAST service [1].
• The agent retrieves 3D models from the AlphaFold Protein Structure Database to assess structural similarity to known proteins [1].
• The agent analyzes the genomic neighborhood, such as checking if a candidate sits near an ABC transporter permease or an ATP-binding protein [1].
• The agent reviews existing literature to propose a mechanism of action and a predicted potency spectrum [1].
• The agent translates computational findings into a bench protocol for protein expression, antimicrobial susceptibility testing, target engagement, and in vitro toxicity testing [1].

Limits, privacy and academic integrity

While Antigravity accelerates the search process, the discovery of viable proteins still requires human intuition and iteration to include new empirical evidence [1]. The transition from computational prediction to functional characterization remains a complex process [2]. Challenges in the broader field of AMP development include high costs for production and development, as well as the need to ensure peptides maintain antibacterial properties while minimizing toxicity [2].