Google DeepMind Creates The First Watermarked AI-Designed Proteins That Still Work
0:00 / 0:39
News
Google DeepMind Creates The First Watermarked AI-Designed Proteins That Still Work
calendar_today Date:
schedule Duration: 0:39
database
Summary Report
DeepMind's SynthID Bio made the first watermarked AI-designed proteins that still work after synthesis.
- 01. Watermark steers amino acid choices via ProteinMPNN and a fine-tuned AlphaFold 3
- 02. Lab-tested binders against spike RBD, VEGF-A and PD-L1 still bound; aimed at biosecurity
Google DeepMind's SynthID Bio subtly steers which amino acids a model chooses, leaving a hidden watermark in the sequence without changing how the protein folds or functions. It works with ProteinMPNN for sequences and a fine-tuned AlphaFold 3 for structures. In the lab, watermarked binders against targets including the SARS-CoV-2 spike protein and PD-L1 still bound, and the watermark survived synthesis. The goal is biosecurity, letting DNA synthesis providers and regulators trace where AI-designed proteins come from.
Meta Data
Company:
Model:
