Track genomics, neuroscience and molecular biology.
Search papers, labs, and topics across Lattice.
This paper introduces GenoTrace, a codon-aware watermarking method designed to embed detectable signals of synthetic training data within distilled genome foundation models. By modulating teacher generation bias based on codon position and organism-specific usage, GenoTrace enables auditability of student models without requiring the original watermark. Experiments demonstrate high detection rates and robustness against adversarial edits, establishing a practical approach for tracing data provenance in genomic AI.
Genomic foundation models can be watermarked through distillation, retaining a detectable signal of synthetic training data even after significant model compression and adversarial attacks.
Can a genome model retain a detectable record of the synthetic sequences used to train it? We study watermark inheritance through distillation with GenoTrace, a codon-aware extension of green-list watermarking. Two token-level factors modulate the teacher's generation bias using codon position and organism-specific codon usage. The resulting sequences train a smaller student, whose outputs are audited without an active watermark processor. In a three-seed GenomeOcean-500M-to-100M experiment, the joint configuration achieves a mean audit score of 17.88 and 94.5% detection at a fixed threshold. It retains 49.0% detection after key-aware token substitution, compared with 0% for the available single-seed plain-watermark comparator, and 47.0% after combined mechanism-targeted nucleotide edits. Additional experiments establish inherited signal across five organism-conditioned datasets and teacher-student size ratios up to 40. Component ablations and computational sequence-quality assays reveal distinct operating points for detection strength and coding coverage. GenoTrace provides a practical token-level construction and an empirical account of how genomic structure shapes inherited watermark signals. The findings concern shared-tokenizer distillation and the tested editing procedures, with calibration and biological utility treated as separate evaluation requirements.