Google DeepMind (this site on google deepmind) (explore this topic) Brings SynthID Watermarking to Synthetic Biology
Google DeepMind has introduced SynthID Bio, a tool that embeds a unique signature into the amino acid sequence or 3D structure of proteins, making it possible to track their provenance. This watermarking technology, designed to enhance biosecurity, was recently demonstrated in a blog post and a paper published in Nature.
How SynthID Bio Works
For amino acid sequences, SynthID Bio subtly alters the choice of acids. In 3D structures, it adjusts atomic coordinates. The tool integrates seamlessly with ProteinMPNN, a popular protein design tool, utilizing a cryptographic key to suggest amino acids. ProteinMPNN ensures the suggested sequences result in functional proteins.
Lab Tests and Results
The team tested their approach with binders, proteins designed to interact with others. They achieved promising results on three targets: VEGF-A, the SARS-CoV-2 spike protein RBD, and PD-L1. The watermarked designs performed similarly to unmarked ones in terms of hit rate, binding affinity, and sequence diversity.
Applications and Benefits
SynthID Bio could significantly aid DNA synthesis companies in screening orders for potential threats. With AI designing sequences that may resemble known hazards, a watermark could verify the origin of a design, streamlining biosecurity processes. The technology also has the potential to label AI-created entries in public databases like the Protein Data Bank, UniProt, and GenBank.
Limitations and Future Work
While SynthID Bio demonstrates promising capabilities, the researchers acknowledge several limitations. The security of the system depends on the protection of its keys during sharing and storage. Very short proteins might not carry enough marked amino acids for reliable detection. Additionally, fusing a marked protein with an unmarked one could dilute the watermark signal. The current approach uses statistical detection, which means setting a cutoff can influence the balance between false positives and negatives.
This breakthrough by Google DeepMind opens new avenues for ensuring the responsible use of AI in synthetic biology, addressing critical concerns about the security and traceability of designed proteins.