scigantic-cryoet

Search the CZ CryoET Data Portal from Python.

import scigantic_cryoet as cryoet

cat = cryoet.CryoetCatalog()
cat.search("Legionella", has_annotations=True)

Installation

$ pip install scigantic-cryoet

Why this exists

s3://cryoet-data-portal-public is public and anonymous, but its top level is nothing but numeric dataset ids. There is no search: finding "a tomogram of a ribosome in a mammalian neuron" otherwise means opening dataset_metadata.json by hand, once per dataset, across all ~370 of them. This package is a catalog and search layer on top of that bucket, the same move scigantic-empiar and scigantic-emdb made for their own archives.

It does not read tomogram pixel data or reimplement OME-Zarr access. The portal already ships a thumbnail.gif/snapshot.gif per dataset (unlike EMPIAR, which ships no previews at all), and the official cryoet-data-portal client already reads the zarr volumes well. This package's job ends at "which dataset, and where."

The catalog

CryoetCatalog loads a pre-built index so search/filter across every dataset is instant, no live reads. Measured against the full portal, 2026-09-02 (370/370 datasets, 0 failures, 129s to build):

Read catalog_meta before trusting a fill rate. Fields drawn from dataset_metadata.json (title, organism, sample_type, disease, assay, cross-references, ...) are COMPLETE: every dataset has one. Fields that live one level down at the run/tomogram level (voxel_spacings, reconstruction_method, ctf_corrected, annotation_objects, ...) describe one run, named in sampled_run: walking every run of every dataset is thousands of listings for detail that barely varies within a dataset. catalog_meta names exactly which fields are sampled, so this can't be missed the way an earlier internal EMPIAR catalog once advertised "all ~3,000 entries" while actually holding 12.

cat = cryoet.CryoetCatalog()
cat.catalog_meta        # {"catalog_entries": 370, "sampled_fields": [...], ...}

cat.search(organism="Homo sapiens", has_annotations=True, sort="runs")
cat.search(has_emdb=True, sort="runs")   # datasets cross-referenced to EMDB, most runs first
cat.gallery(cat.search("spike"))         # HTML gallery, portal thumbnails

The index isn't published yet. CryoetCatalog() with no argument points at where the monorepo's onboarding batch job will land it. Until then, point it at a local file built with the catalog script:

cat = cryoet.CryoetCatalog(url="/path/to/cryoet-catalog.json")
# or: export SCIGANTIC_CRYOET_CATALOG=/path/to/cryoet-catalog.json

Live reads

CryoetClient hits the portal's bucket directly, for a dataset newer than whatever snapshot is loaded, or to double-check a stale field:

client = cryoet.CryoetClient()
client.dataset_metadata(10000)   # full dataset_metadata.json, live
client.runs(10000)               # run directory names, COMPLETE (not sampled)

Data license

The portal describes its data as "publicly available" / "open access" (checked 2026-09-02, portal homepage and Terms of Use), but does not state a specific license designation (CC0, CC-BY, or otherwise) on either page. Check a dataset's own citation/attribution requirements (via dataset_metadata.json's authors/publications fields, or the portal's own dataset page) before redistributing.

What this doesn't do (yet)

No synonym/alias expansion in search() (a plain "GPCR" won't match "G protein-coupled receptor"); scigantic-empiar's query expansion is the template once this has real query traffic to tune against. No walk of every run per dataset (see the fill-rate table above); if you need every run's metadata for a specific dataset of interest, use CryoetClient.runs() plus the official cryoet-data-portal client.