Editorial disclosure

AI assisted primary-source inspection, evidence synthesis and drafting. The user selected the slate and authorized publication. This preparation does not claim completed human source verification; supplements, raw data and code were not independently audited or reproduced.

What the paper claims

The authors report function-preserving watermarking in designed protein binders and an imperceptible structure watermark from a fine-tuned AlphaFold 3 model. These are proof-of-concept provenance signals, not tamper-proof histories.

What was tested

Sequence watermarks were embedded during ProteinMPNN sampling. Watermarked and non-watermarked binders were experimentally compared for affinity and hit rate. A separate fine-tuned AlphaFold 3 approach was evaluated computationally for structural watermark detection and prediction quality.

What the evidence supports

SynthIDBio combines sequence watermarking during ProteinMPNN sampling with a separate structure-watermarking approach based on fine-tuned AlphaFold 3. Watermarked and non-watermarked binders were experimentally compared across three targets; the authors report comparable binding-affinity distributions and hit rates. Structure tests evaluate detectability and prediction quality computationally. This is a technical proof of concept, not a deployed screening system.

Limitations

Watermarks can lose detectability after downstream modifications; robustness and detection trade-offs limit deployment. Both approaches encode watermark presence rather than a detailed provenance history, and detection relies on restricted keys/models. Tested binder function does not establish general functional preservation, biological safety or therapeutic efficacy.

Open-access abstract, main-text evidence, limitations and data/code statements read. Authors link sequence code and in vitro data, plus instructions for access to structure-model weights, at GitHub. Repository contents, weight-access conditions and supplements were not independently audited; no replication or deployment validation was performed. Evidence scope: AI inspection of the abstract and selected full-text sections; no independent replication, exhaustive supplementary review or completed human source verification is claimed.

Sources

Corrections & versioning

Current version 1. Initial evidence brief with explicit source scope, limitations and missing information.

No later corrections logged.