A Labeled Dataset for AI-based Cryo-EM Map Enhancement

An open-source dataset for cryo-EM density map denoising comprising 650 high-resolution experimental maps paired with three types of generated label maps: regression maps capturing idealized density distributions, binary classification maps distinguishing structural elements from background, and atom-type classification maps.

Dataset Download

To keep the data files of denoise dataset permanent, we published all data to the Harvard Dataverse (https://doi.org/10.7910/DVN/CI0J2B), an online data management and sharing platform with a permanent Digital Object Identifier number for each dataset.

Description of the dataset

Dataset Link : https://doi.org/10.7910/DVN/CI0J2B .

The dataset can be accessed using the above download link. The dataset follows the format described below, and please read the subsequent section describing what these data mean.

│── 6bco
    │-- 6bco.mrc
    |-- 6bco.pdb1
    |-- 6bco_classification_situs.mrc
    |-- 6bco_classification_types_situs.mrc
    |-- 6bco_regression_situs.mrc
    |-- 6bco_situs_simulated.mrc

│── 7ki6
    │-- 7ki6.mrc
    |-- 7ki6.pdb1
    |-- 7ki6_classification_situs.mrc
    |-- 7ki6_classification_types_situs.mrc
    |-- 7ki6_regression_situs.mrc
    |-- 7ki6_situs_simulated.mrc

│── 7o3h
    │-- 7o3h.mrc
    |-- 7o3h.pdb1
    |-- 7o3h_classification_situs.mrc
    |-- 7o3h_classification_types_situs.mrc
    |-- .
    |-- .

.
.
.

As shown in the example data format above, each individual directory for the cryo-EM density map (directory name is the protein PDB Code) provides the following data files:

• 6bco.mrc : The deposited experimental cryo-EM density map. The filename is the PDB code of the cryo-EM density map.
• 6bco.pdb1 : The atomic biological assembly of the macromolecule.
• 6bco_situs_simulated.mrc : The simulated cryo-EM density map of 6bco.pdb1 generated using Situs package.
• 6bco_classification_situs.mrc : The classification label map, neighboring voxels are assigned a value of 2, central atomic positions (value 1) and the background (value 0).
• 6bco_classification_types_situs.mrc : The atom type classification label map, values to voxels based on the corresponding atom type: Cα (1), Cβ (2), carbonyl carbon (3), oxygen (4), and nitrogen (5).
• 6bco_regression_situs.mrc : This map contains continuous density values derived from the simulated cryo-EM density map. At each voxel position corresponding to an atom in the PDB structure, we assign the density value from the simulated map, providing a noise-free target for regression-based learning.

Programs to generate the dataset

python3 generate_labels_all_conv.py

Evaluating map quality with Phenix mtriage

The below runs phenix.mtriage to compute the FSC scores. The webpage of Phenix mtriage is available here: https://phenix-online.org/documentation/reference/mtriage.html

python3 evaluate_phenix_mtriage_exp.py
python3 evaluate_phenix_mtriage_reg.py
python3 evaluate_phenix_mtriage_sim.py

Programs to extract the information from Phenix mtriage output

The FSC scores reported in the paper are generated using Phenix mtriage. Use the below to extract information in .csv format

python3 parse_get_resolution_from_log_file_exp.py
python3 parse_get_resolution_from_log_file_reg.py
python3 parse_get_resolution_from_log_file_sim.py

Contact Information

If you have any question, feel free to open an issue or reach out to us: ngzvh@missouri.edu, chengji@missouri.edu.

Citing this work

If you use the code or data in this package, please cite:

@article{giri2025labeled,
  title={A Labeled Dataset for AI-based Cryo-EM Map Enhancement},
  author={Giri, Nabin and Chen, Xiao and Wang, Liguo and Cheng, Jianlin},
  journal={Computational and Structural Biotechnology Journal},
  year={2025},
  publisher={Elsevier}
}