StrucTrace: Fourier Watermarking for Traceable Bio-molecular Assets

Abstract

The rise of generative artificial intelligence (GenAI) in protein and nucleic acid design has created unprecedented opportunities for synthetic biology, but also heightened the need for reliable provenance and intellectual-property protection. To meet this challenge, we present a Fourier domain watermarking framework that encodes digital identifiers directly into three-dimensional biomolecular structures. By perturbing only flexible backbone atoms and embedding information through frequency domain modulation, the method achieves imperceptible alterations while ensuring deterministic and reversible decoding. Large-scale validation on over 40,000 protein structures demonstrates its robustness: structural deviations remain orders of magnitude below biological thresholds, watermarks are recovered with perfect accuracy, and functional analyses confirm stability at both thermodynamic and dynamic levels. Beyond technical performance, the approach provides a foundation for a broader ecosystem of secure biomolecular asset management, integrating provenance verification, access control, and digital rights management. Together, these advances establish biomolecules as traceable and auditable digital assets, aligning the future of bio-design with emerging standards for trustworthy AI.