CyteBase finds and opens published datasets; CyteArc analyzes them. This quick example mounts a small dataset as a local, writable CyteArc store whose counts stay in the bucket, runs CyteArc’s RNA pipeline on it, and plots the result. Plotting needs CyteArc’s plotting dependencies, described in Plotting. The whole example takes a few minutes for the 1,790-cell dataset it selects.
Choose a dataset¶
import cytearc
import cytebase
# Keep routine pipeline progress out of the output.
cytearc.configure_output(level="WARNING", progress=False)
# Pin the October 2026 release and read it anonymously.
catalog = cytebase.Catalog(version="2610", token=False)
# List the smallest ready human datasets with at least 1,500 cells.
small = catalog.query(
"SELECT cytebase_id, title, cell_count FROM datasets "
"WHERE status = 'ready' AND processed_version_id = latest_version_id "
"AND list_has_any(organism_labels, ['Homo sapiens']) AND cell_count >= 1500 "
"ORDER BY cell_count, cytebase_id LIMIT 5"
)
dataset_id = small[0]["cytebase_id"]In release 2610, the first row is mcglade_2021_scrna_data_human_fallopian_tube_30f5e171, a
fallopian tube dataset with 1,790 cells. Explore CyteBase shows the other ways to
search the catalog.
Mount the dataset and run the pipeline¶
# Create a writable local store; its counts stay in the bucket.
analysis = catalog.mount_datastore(dataset_id, at="quickstart.zarr")
# Run CyteArc's RNA pipeline; every result is saved in quickstart.zarr.
run = analysis.pipeline.run(label="baseline")
# Color the computed UMAP by the selected clustering.
analysis.plots.embedding(run=run, color_by="clusters")The pipeline filters cells, selects highly variable genes, and computes PCA, a neighbour graph, a UMAP, Leiden and Paris clusterings, doublet scores, and marker genes. It reads count blocks from the bucket as it needs them, so it runs wherever your Python session runs.
Read the results¶
# Show the three strongest markers of each cluster.
markers = analysis.markers.load(marker=run["markers"])
markers.groupby("group_id", sort=False).head(3)[["group_id", "feature_name"]]The mount keeps your results. Reopen it in a later session with
catalog.mount_datastore(dataset_id, at="quickstart.zarr"), which checks it against the same
published build, or with cytearc.DataStore("quickstart.zarr", min_features_per_cell=-1).
Next steps¶
CyteArc’s scRNA-seq tutorial explains each pipeline step and how to choose its settings.
Joint reanalysis in CyteArc combines several CyteBase datasets in one analysis without copying their counts.
CyteBase case study: COVID-19 blood compares donors across conditions using published annotations.