Google DeepMind announced the introduction of SynthID Bio, a watermarking technology built to embed an imperceptible signature directly into biological code. The system works on digital models as well as synthesized, physical proteins while preserving biological function during laboratory testing.
Generative artificial intelligence helps scientists address biological challenges by predicting protein structures with AlphaFold and designing new proteins using AlphaProteo and ProteinMPNN. However, these tools also introduce new risks, including novel designs that bypass traditional DNA synthesis screening and mislabeled 3D structures that can pollute public databases.
Watermarking proteins
SynthID Bio adapts its approach based on data types, guiding amino acid choices for sequences and adjusting atomic coordinates for predicted 3D structures to create a reliable detection signal. In experiments, these adjustments maintained the biological function required for treating disease and advancing research.
Google DeepMind tested the approach on protein binders built using AlphaProteo alongside a SynthID Bio-enabled version of ProteinMPNN. In wet-lab testing across VEGF-A, the SARS-CoV-2 spike protein RBD, and PD-L1, the watermarked designs matched the hit rate, binding affinity, and natural sequence diversity of unwatermarked versions.
For protein folding, SynthID Bio fine-tunes a small part of AlphaFold 3's diffusion network to build watermarking directly into model weights. The method preserves prediction accuracy while offering near-detectability and holding up against digital noise or minor coordinate changes.
Biosecurity applications
Sarah Carter, a biosecurity policy expert and Principal at Science Policy Consulting, reviewed the work and stated that the watermarks empower developers to lead on safety and allow synthesis providers to streamline screening. James Diggans, Vice President of Policy and Biosecurity at Twist Bioscience, provided early feedback on the paper.
Google DeepMind collaborated with the Hie lab at Stanford University and the Arc Institute to integrate SynthID Bio into Evo 2, an advanced genomic model, to watermark the genome of an engineered bacteriophage. Early laboratory testing in bacteria cultures confirmed those watermarked bacteriophages are functional, and researchers plan to share more details in a technical manuscript.
