MolMark: Safeguarding Molecular Structures with Atom-Level Watermarking

Official implementation of [MolMark: Safeguarding Molecular Structures with Atom-Level Watermarking]

Update

This work is based on the implementation of [Unified Generative Modeling of 3D Molecules with Bayesian Flow Networks] (https://github.com/AlgoMole/GeoBFN)

Related works

(1) FoldMark: Safeguarding Protein Structure Generative Models with Distributional and Evolutionary Watermarking (https://github.com/zaixizhang/FoldMark)

(2) Securing the Language of Life: Inheritable Watermarks from DNA Language Models to Proteins

Prerequisite

You will need to have a host machine with gpu, and have a docker with nvidia-container-runtime enabled.

Overview

MolMark is the first watermarking strategy designed to protect moleucles. It:

• Maintains Molecular Properties: Operates at the atom level, embedding watermarks by subtly modulating chemically informed features.
• Ensure Molecular Functionlaity: Guarantes the physicochemical properties and the functionality in docking performance.
• Exhibits High Bit Accuracy: Achieves over 95% watermark bit accuracy at 16 bits with minimal impact on structural integrity.
• Presents Robust Against SE(3) Transformations: Has high robustness against rotation, translation, and reflection with bit accuracy higher than 90%.

Results

Application scenarios and structures of MolMark in protecting molecules

Structure of eight pairs of molecules

Quick start

Environment setup

Clone the repo with git clone,

git clone https://github.com/RunwenHU/MolMark.git

Train a model on qm9 dataset

python main.py --config_file configs/bfn4molgen.yaml --epochs 3000 --no_wandb

Citation

If you find this work helpful, please cite our paper:

@article{hu2024molmark,
  title={MolMark: Safeguarding Molecular Structures with Atom-Level Watermarking},
  author={Hu, Runen and Chen, Peilin and Ding, Keyan and Wang, Shiqi},
  journal={arXiv},
  year={2025},
}