CyteBase

A searchable catalog of public single-cell datasets, stored as cloud-hosted CyteArc DataStores.

Search by study, tissue, disease, or organism, then open a dataset from a local notebook or a cloud Python session. Explore its published cell annotations, imported embeddings, and gene expression without downloading an H5AD or a complete store. Public access needs no credentials.

Install CyteBase · Explore the catalog · API reference

What the catalog contains¶

The datasets come from CZ CELLxGENE Discover. Each public release is a separate, read-only Hugging Face bucket named by year and month. The October 2026 release (2610, in Nygen/cytebase-2610) was built from a CELLxGENE Discover snapshot taken on 2026-10-10 and contains primary RNA datasets from 376 registration-ready collections, converted to CyteArc stores. cytebase.releases() lists every release with its bucket, notes, and changelog.

Each catalog entry records the dataset’s title, citation, CELLxGENE links, size, and ontology labels for its tissues, diseases, cell types, assays, and organisms. The stores hold the RNA counts with the published cell annotations. When a dataset provides a usable UMAP (obsm/X_umap), the build imports it, so it can be plotted without recomputing an embedding. Cite the original study and CELLxGENE Discover when you use these data.

A first example¶

import cytebase

# Connect to the latest public release; no credentials are needed.
catalog = cytebase.Catalog()

# Search IDs, titles, citations, authors, and ontology labels.
matches = catalog.search("kidney")
print(matches)

# Open one of the matching datasets for read-only exploration.
dataset_id = matches[0]["cytebase_id"]
ds = catalog.open_datastore(dataset_id)

# Plot the imported UMAP; plotting needs CyteArc's extra dependencies.
umap_ref = cytebase.embedding(ds, "X_umap")
ds.plots.embedding(layout=umap_ref, color_by="cell_type")

# Create a writable local store for your own analysis, keeping counts remote.
analysis = catalog.mount_datastore(dataset_id, at="kidney_analysis.zarr")

How it works¶

• A local, verified catalog. Catalog() keeps a small DuckDB copy of the catalog on your machine and checks it against the published SHA-256 checksum before each query. Searching never reads count data.

• Remote DataStores. open_datastore validates the published build and returns a read-only CyteArc DataStore. CyteArc reads metadata and count blocks over the network as needed, and computation runs where your Python session runs.

• Writable mounts. mount_datastore creates a local CyteArc store for new analysis results while the counts stay in the bucket. Use it with CyteArc’s analysis pipeline.

• Reproducible releases. Pin a release with Catalog(version="2610") to reproduce discovery results; newer releases may return different matches.

Next steps¶

• Installation covers the Python requirements, plotting dependencies, releases, credentials, and network access.

• Explore CyteBase searches the catalog, reads annotations, plots the imported UMAP and gene expression, and queries the catalog with SQL.

• Analyze a dataset with CyteArc mounts a small dataset and runs CyteArc’s RNA pipeline on it.

• CyteBase case study: COVID-19 blood compares COVID-19 and healthy blood with donor-level summaries.

• Joint reanalysis of CyteBase datasets in the CyteArc documentation combines several CyteBase datasets into one analysis.

• The CyteArc documentation describes the analyses you can run on an opened or mounted store.

• The API reference lists every public function and class.

• Build pipeline describes how operators build and publish releases.

Report problems in the issue tracker. The documentation index for LLMs lists guides and API references for assistants that read this site.