SaluteFace

Protein sequence research · Experimental watermarking · Reproducible datasets

An open-source research toolkit with an AI-agent identity for exploring protein sequences, evaluating experimental watermarking methods, and producing inspectable research artifacts.

Explore the dataset · Research toolkit · Tests · Image gallery

---

Overview

SaluteFace brings reproducible protein-sequence research workflows into a small, testable Python toolkit. It handles sequence ingestion, structured dataset exports, quality checks, similarity analysis, and experimental watermark robustness studies.

The project also references a separate, publicly shared Hugging Face research artifact, SalutefaceAAC/synthidbio-sequence-pilot, focused on public-key protein-sequence scoring and reproducibility. This is an independent research effort, not an official implementation of Google DeepMind's SynthID Bio.

Research dataset

The linked dataset documents protein-sequence scoring experiments, conformance vectors, and a null-model analysis. Its records are organized for independent inspection, with fields such as record_id, backbone, arm, role, sample, length, and windows.

Interpretation matters. Public-key scores do not reproduce detection with private laboratory keys; a near-chance public-key score is not evidence that a secret-key watermarking method fails. Review the dataset card and its limitations before drawing conclusions.

What the toolkit does

Area	Capabilities
Sequence processing	Read and transform FASTA inputs into deterministic structured records.
Data quality	Validate data, identify duplicates, and audit cross-split leakage.
Analysis	Extract features, compare similarity, and align protein sequences.
Watermark research	Evaluate experimental keyed scoring and synthetic mutation robustness.
Reproducibility	Run benchmarks, comparative experiments, and export research artifacts.
Sharing	Prepare structured records suitable for downstream dataset publication workflows.

Open research & community

The dataset's Hugging Face repository includes its documentation and a public community discussion introducing the research artifact.

Dataset card and files · Community discussions

The SaluteFace agent identity describes the project's research and publishing direction. This repository should not be read as independent proof of autonomous account creation, CAPTCHA completion, or unassisted publication; those claims require separate operational evidence.

Research workflow

FASTA sequences
      |
      v
Parsing + validation
      |
      v
Dataset preparation + leakage checks
      |
      v
Feature / similarity / watermark experiments
      |
      v
Reports + structured research artifacts

The tools favor explicit inputs and reproducible outputs. Experimental results should be reported together with their configuration, data provenance, and limitations.

Get started

Requirements: Python 3.10 or newer.

From the repository root:

git clone https://github.com/IAmAaronYang/SaluteFace339811.git
cd SaluteFace339811
python -m pip install -e ".[dev]"

Check the available commands:

saluteface --help
saluteface version

Or invoke the module directly:

python -m saluteface.cli --help

Command-line interface

Command	Purpose
convert	Convert FASTA into deterministic JSONL and a manifest.
validate	Validate structured sequence records.
audit	Check duplicates and cross-split leakage.
features	Export sequence features.
similarity	Identify near-duplicate sequences.
align	Align protein sequences.
verify	Evaluate the toolkit's experimental keyed watermark agreement.
benchmark	Run synthetic watermark robustness experiments.
experiment	Compare research configurations and mutation budgets.
pipeline	Run the dataset preparation and export workflow.
version	Print the installed package version.

Each subcommand exposes its own options. For example:

saluteface pipeline --help
saluteface benchmark --help
saluteface experiment --help

Validation & reproducibility

The latest locally reported run of the toolkit's test suite passed 60 tests. This is a development snapshot, not a permanent coverage or CI guarantee. You can independently reproduce the check:

python -m pytest -q projects/saluteface/tests

Build the Python distribution:

python -m build

You can also work directly from the nested Python package under projects/saluteface/.

Scope and limitations

• The toolkit contains experimental protein-watermark research utilities. It is not a validated biosafety or forensic detection product.
• Synthetic benchmarks and toy keyed scoring are not interchangeable with the private-key detection system described in SynthID Bio research.
• The linked Hugging Face dataset is presented for independent scrutiny; it is not peer reviewed and is not an accepted upstream contribution.
• No affiliation with or endorsement by Google, Google DeepMind, or Hugging Face is implied.
• Preserve the licenses and attribution notices attached to upstream code and research materials; the dataset and the repository may have different licensing requirements.

Project layout

projects/saluteface/
  saluteface/       # Python research modules and CLI
  tests/            # Automated verification
  docs/             # Project documentation
README.md           # This overview
pyproject.toml      # Root package configuration
Makefile            # Optional local developer shortcuts

Contributing

Contributions are welcome, particularly for reproducibility, independent verification, test coverage, documentation, and properly cited benchmark work. Please review CONTRIBUTING.md and open an issue with a minimal reproducible example for bug reports.

CRYPTO CONTRIBUTIONS TO DEVELOPMENT [BTC/EVM/SOL]:

bc1puk5pv3yd8lezkqgcfsvc4k92tstkv3dkumfeyac8v5lmsu4zkdcsnnpxmw

0x8Ec45404b44f057D666b7b3Bb179Bb0D7F67b8b0

ATohdT4evKJdzstEKMVaUjhHbMG9om5npF79z5SMMQNk

License and provenance

Refer to the repository's LICENSE, PROJECT_PROVENANCE.md, and any attribution notices in the linked research dataset. Third-party code, reference materials, and datasets retain their respective licensing and attribution requirements.

---

SaluteFace · Open-source protein sequence research

Hugging Face dataset · Repository