Skip to content

Evaluation of Aldolase Micrographs from EMPIAR-10379

This tutorial shows how to analyze 1,118 motion corrected micrographs of rabbit muscle aldolase from EMPIAR-10379.

1. Prepare Input Data

First, we need to create a metadata table used for prismPYP training and embedding generation containing information about microscope parameters, CTF statistics, and motion information for all micrographs in the dataset.

In general, you can build metadata using either nextPYP preprocessing outputs or cryoSPARC outputs, but for this example we will use pre-calculated results.

🧪 Download the Test Data

We will download the example_data.tar.gz archive from Zenodo, which contains micrograph and power-spectrum images plus all the necessary metadata:

mkdir example_data
tar -xvzf example_data.tar.gz -C example_data

This command extracts the data into an example_data/ folder containing:

example_data/
    ├── pkl/                      # metadata from nextPYP preprocessing
    ├── webp/                     # 512×512 images (micrographs + power spectra)
    ├── J7_exposures_accepted_exported.cs
    ├── sp-preprocessing-*.micrographs
    └── .pyp_config.toml

📦 Intermediate Results

The Zenodo entry also contains the following files:

  • model_weights.tar.gz: Trained model weights for the real domain (real_model_best.pth.tar) and the Fourier domain (fft_model_best.pth.tar) inputs.
  • fft_good_export.parquet: Data points that have high-quality features in the Fourier domain.
  • real_good_export.parquet: Data points that have high-qualtiy features in the real domain.

By taking the intersection between fft_good_export.parquet and real_good_export.parquet, you can obtain the 862 high-quality micrographs that we used to obtain a 2.9 Å structure of aldolase.

2. Build Metadata Table

Before starting, create directories to store the metadata and all generated outputs:

mkdir -p metadata
mkdir -p output_dir
  • Run the following command to assemble metadata from nextPYP preprocessing results: 
    prismpyp metadata_nextpyp \
   --pkl-path example_data/pkl \
   --output-dir metadata \
   --cryosparc-path example_data/J7_exposures_accepted_exported.cs

You can omit --cryosparc-path if you do not need relative ice thickness visualization.

To build metadata directly from cryoSPARC outputs, you’ll need data from the Import, Manually Curate Exposures (with outputs of Patch CTF Estimation as the inputted micrographs), and CTFFIND4 jobs.

For the test dataset (EMPIAR-10379), the deposited data already contains motion corrected micrographs, so you can skip motion correction.

  • Export the outputs of the following jobs and note their locations:

    • Import Micrographs → J1
    • Patch CTF Estimation → J2
    • CTFFIND4 → J3
    • **Manually Curate Exposures→J4`
    • cryoSPARC project directory → /cryosparc/output/dir

  • Build the metadata table: 

    prismpyp metadata_cryosparc \
   --imported-dir "/cryosparc/output/dir/J1/imported" \
   --patch-ctf-file "/cryosparc/output/dir/exports/groups/J4_exposures_accepted/J4_exposures_accepted_exported.cs" \
   --ctffind-dir "/cryosparc/output/dir/J3/ctffind_output" \
   --ctffind-file "/cryosparc/output/dir/exports/groups J3_exposures_success/J3_exposures_success_exported.cs" \
   --output-dir metadata

Depending on how many micrographs you have, this process may take several minutes to run.

3. Generated Outputs

The metadata-building command will produce a file named micrograph_metadata.csv, containing:

Column Description
micrograph_name Name of each micrograph
rel_ice_thickness Relative ice thickness (if --cryosparc-path was provided)
ctf_fit CTF fit correlation coefficient
est_resolution Estimated resolution in Å
avg_motion Average beam-induced motion
num_particles Number of picked particles
mean_defocus Mean defocus (Å)

In addition, the following files are generated:

  • pixel_size.txt — microscope pixel size for this dataset
  • all_micrographs_list.micrographs — list of all micrographs (without extensions)
  • webp/ — directory of .webp images for both micrographs and their CTFFIND4-derived power spectra