Xu-AI4Science/CryoDynaScape-data

CryoDynaScape: data and results

Data behind the CryoDynaScape manuscript (network-free, hierarchical cryo-EM heterogeneity analysis): every result, model checkpoint and preprocessed input used for the tables and figures. The code will be released on publication.

Reproduction

Step-by-step instructions for every figure and table: docs/REPRODUCE.md. Software environments: envs/. Exact arguments of every run: logs/ (and the args stored in each checkpoint). The latent behind every AU-FSC evaluation: results/official/latent_sources.json.

Layout

path	content
results/	all runs: bead models, per-particle labels, hierarchy decisions, metrics (metrics.json), latents (*_test_results.h5), official AU-FSC evaluations (results/official/<dataset>/score_ext_*.json, per-query values, generated maps), figure data (results/figures/)
checkpoints/	trained CryoDynaScape models per run
baselines/inputs/	baseline inputs prepared from the benchmark datasets (particle stacks, poses, CTF, STAR files)
baselines/runs/	outputs of the published baseline methods (cryoDRGN, RECOVAR, CryoSTAR, cryoSPHERE, DynaMight, e2gmm)
data/synth_*	synthetic benchmark datasets generated for this work
data/Spike-MD-128/	Spike-MD downsampled to 128 px with ground-truth maps of the query frames
data/structures/	atomic models used (PDB 1HZH, 1N0X, 3AVE, 4YBB; AlphaFold3 predictions; IgG templates)
logs/, envs/, notes/, docs/	run logs, environment specs, data notes, reproduction guide
data/real/empiar10180_128/, data/real/empiar10076_128/	EMPIAR-10180 and EMPIAR-10076 particles downsampled to 128 px with consensus poses and CTF

Sources and licences of third-party data

Derived from CryoBench (Jeon et al. 2024; Zenodo 11629428, 12528292, 14941494; CC BY 4.0), EMPIAR-10180 (Plaschka et al. 2017) and EMPIAR-10076 (Davis et al. 2016), with poses/CTF from the cryoDRGN EMPIAR inputs (Zhong et al.). Full-size original datasets are not re-distributed here; obtain them from those sources. Our own data are released under CC BY 4.0.