AFM-Fold

November 26, 2025 · View on GitHub

AFM-Fold is an implementation of protein structure prediction from AFM (Atomic Force Microscopy) images.

Example Figure

Installation

We recommend using Python 3.10–3.12 with CUDA-enabled PyTorch (tested with PyTorch 2.3.1).
Clone this repository and install dependencies as follows:

# Clone this repository
git clone https://github.com/matsunagalab/afmfold.git
cd afmfold

# Unzip the released data
wget https://zenodo.org/records/17714204/files/results.zip
unzip results.zip
ls results

# Install dependancies
pip install -e .[e2cnn]

Usage

Reproducing the Paper Results

The basic usage is demonstrated in notebooks/example.ipynb. This notebook demonstrates the reproduction of the main results from the paper:

  • (A) Conditional structure generation
  • (B) Evaluation of estimation error using MD data
  • (C) Noise robustness
  • (D) Comparison with rigid-body fitting
  • (E) Guidance scheduling
  • (F) Overviewing training data

Training & Inference

The procedures for training, inference and rigid-body fitting are described in detail in scripts/SCRIPTS.md.

Citation information

If you use AFM-Fold in your work, please cite as follows:

@article{kawai2025afmfold,
  title   = {AFM-Fold: Rapid Reconstruction of Protein Conformations from AFM Images},
  author  = {Tsuyoshi, Kawai and Yasuhiro, Matsunaga},
  journal = {bioRxiv},
  year    = {2025},
  url     = {https://www.biorxiv.org/content/10.1101/2025.11.17.688836v1},
  doi     = {https://doi.org/10.1101/2025.11.17.688836}
}

Acknowledgements

AFM-Fold relies heavily on the implementation of:

License

AFM-Fold is released under the MIT License.
See the LICENSE file for more details.

Contact Us

If you have any questions, issues, or suggestions, please: