MARCO: The Radioactive Watermark for Protein Generative Models
MARCO watermarks protein generative models so marks survive extraction and can trace misused designs.
MARCO is a conformation watermark for protein generative models intended to protect model intellectual property and support forensic tracing of misuse. It embeds the mark during diffusion reverse denoising through an auxiliary encoder-decoder while leaving the original model parameters frozen. Training uses losses on C-alpha pairwise distances and torsion angles inside an adversarial setup with simulated attacks. Experiments report preserved structural fidelity and transfer of the watermark into outputs of models trained on watermarked data.
- MARCO is a radioactive watermark for protein generative models.
- It targets model-extraction defense and forensic misuse traceability.
- Marks are embedded during diffusion denoising with frozen base weights.
- Watermarks transfer into models trained on watermarked outputs.
- Products
- MARCO
From the original article
1 min read at arxiv.orgProtein Generative Models (PGMs) have revolutionized structural biology by enabling the design of complex 3D protein structures from sequence data. However, this breakthrough introduces a dual-use challenge, exposing high-value PGMs to economic risks like unauthorized model extraction and biosecurity threats such as biohazard synthesis.
By Huajie Chen · arxiv.org
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.