Google DeepMind introduces SynthID Bio, watermarking AI-designed proteins without disrupting function

Google DeepMind releases SynthID Bio, a watermarking method for AI-designed proteins, and validates in wet-lab experiments that it does not impair protein function. The method hides a signature in the designed amino acid sequence and in the atomic coordinates of the predicted three-dimensional structure, and DeepMind says the signature can also be detected in the synthesized physical protein. The team paired the AlphaProteo design method with a ProteinMPNN sequence generator with SynthID Bio, and ran wet-lab experiments on three targets: VEGF-A, the SARS-CoV-2 spike protein receptor-binding domain, and PD-L1. The watermarked designs matched the unwatermarked versions in hit rate, binding affinity, and natural sequence diversity.