KokosDev/single-cell-lung-zarr

Single-cell lung (CellxGene Census) — Zarr

This dataset was exported from the CellxGene Census as a chunked + compressed Zarr store intended for easy streaming access.

• Source: CellxGene Census API
• Organism: Homo sapiens
• Filter: tissue_general == 'lung' and is_primary_data == True
• Shape: 100,000 cells × 61,497 genes
• Zarr path: lung.zarr

Compression

• Uncompressed (dense float32): 22.91 GB
• Compressed Zarr: ~307 MB (322 MB on Hub)
• Compression ratio: ~76× (Blosc zstd on sparse single-cell data)

The high compression ratio is achieved through Blosc zstd compression optimized for sparse expression matrices typical in single-cell RNA-seq data.

Included labels

• obs/cell_type
• obs/disease
• obs/tissue
• obs/sex
• obs/assay

Cell type distribution (top 25)

cell_type	n_cells
neuron	8,037
oligodendrocyte	4,890
fibroblast	4,539
glutamatergic neuron	3,367
macrophage	2,216
endothelial cell	1,941
astrocyte	1,808
natural killer cell	1,803
T cell	1,734
malignant cell	1,625
GABAergic neuron	1,589
classical monocyte	1,562
basal cell of prostate epithelium	1,401
myeloid cell	1,379
plasma cell	1,263
epithelial cell of proximal tubule	1,185
adipocyte of omentum tissue	1,170
microglial cell	1,164
acinar cell of salivary gland	1,074
oligodendrocyte precursor cell	1,044
monocyte	1,036
blood vessel endothelial cell	982
pericyte	942
subcutaneous adipocyte	932
CD8-positive, alpha-beta T cell	830

Disease labels included

Total unique disease labels: 72

Alzheimer disease
B-cell acute lymphoblastic leukemia
Barrett esophagus
COVID-19
Crohn disease
Down syndrome
HIV infectious disease
HIV infectious disease || leishmaniasis
HIV infectious disease || visceral leishmaniasis
Lewy body dementia
Parkinson disease
Wilms tumor
acute myeloid leukemia
acute promyelocytic leukemia
adenocarcinoma
age related macular degeneration 7
anencephaly
arrhythmogenic right ventricular cardiomyopathy
atrial fibrillation || mitral valve insufficiency
basal cell carcinoma
basal laminar drusen
benign prostatic hyperplasia
breast cancer
breast carcinoma
cataract
chromophobe renal cell carcinoma
clear cell renal carcinoma
colon adenocarcinoma
colon sessile serrated adenoma/polyp
colorectal cancer
common variable immunodeficiency
cytomegalovirus infection
dementia
dilated cardiomyopathy
enamel caries
gastric intestinal metaplasia
gastritis
glioblastoma
hyperplastic polyp
hypertrophic cardiomyopathy
influenza
invasive ductal breast carcinoma
invasive lobular breast carcinoma
leukoencephalopathy, diffuse hereditary, with spheroids 1
long COVID-19
luminal A breast carcinoma
luminal B breast carcinoma
macular degeneration
malignant ovarian serous tumor
multiple sclerosis
myocardial infarction
myocarditis
neuroblastoma
neuroendocrine carcinoma
non-compaction cardiomyopathy
normal
obstructive nephropathy
oral cavity squamous cell carcinoma
pilocytic astrocytoma
plasma cell myeloma
prediabetes syndrome
prostatic acinar adenocarcinoma
pulmonary emphysema
pulpitis
respiratory failure
rheumatoid arthritis
temporal lobe epilepsy
triple-negative breast carcinoma
tubular adenoma
tubulovillous adenoma
type 2 diabetes mellitus
ulcerative colitis

Data layout

• Expression matrix: X (dense 2D array)
  • chunks: (1000, 1000)
  • dtype: float32
  • compression: Blosc zstd
• Metadata:
  • obs/_index, obs/<col> (or obs/<col>_codes + obs/<col>_categories for categoricals)
  • var/_index, var/<col> (or var/<col>_codes + var/<col>_categories for categoricals)

Loading (zarr → AnnData)

import zarr
import numpy as np
import pandas as pd
from anndata import AnnData

root = zarr.open_group("lung.zarr", mode="r")
X = root["X"]  # zarr Array (lazy / on-demand)

obs = pd.DataFrame(index=root["obs/_index"][:].astype(str))
for col in ["cell_type", "disease", "tissue", "sex", "assay"]:
    codes_key = "obs/" + col + "_codes"
    cats_key = "obs/" + col + "_categories"
    if codes_key in root and cats_key in root:
        obs[col] = pd.Categorical.from_codes(root[codes_key][:], root[cats_key][:].astype(str))
    elif "obs/" + col in root:
        obs[col] = root["obs/" + col][:].astype(str)

var = pd.DataFrame(index=root["var/_index"][:].astype(str))
for col in ["feature_name", "feature_id", "gene_symbol"]:
    key = "var/" + col
    if key in root:
        var[col] = root[key][:].astype(str)

# Convert to in-memory AnnData (loads full matrix):
adata = AnnData(X=np.asarray(X), obs=obs, var=var)

Scanpy workflow example

import scanpy as sc

sc.pp.normalize_total(adata, target_sum=1e4)
sc.pp.log1p(adata)
sc.pp.highly_variable_genes(adata, n_top_genes=2000, flavor="seurat_v3")
adata = adata[:, adata.var["highly_variable"]].copy()

sc.pp.scale(adata, max_value=10)
sc.tl.pca(adata, svd_solver="arpack")
sc.pp.neighbors(adata, n_neighbors=15, n_pcs=50)
sc.tl.umap(adata)
sc.tl.leiden(adata, resolution=0.5)

Streaming benchmark (single 1000×1000 chunk)

Measured locally while building this dataset:

• open Zarr group: 0.0021 s
• read one X[chunk] (1000×1000): 0.0864 s

{
  "open_seconds": 0.0021242350339889526,
  "read_chunk_seconds": 0.08641284704208374,
  "chunk_shape": [
    1000,
    1000
  ],
  "chunk_origin": [
    84212,
    38534
  ],
  "shape": [
    100000,
    61497
  ]
}

Build metadata

{
  "census_version": "2025-11-08",
  "created_at": "2026-02-03T06:49:50.898169+00:00",
  "max_cells": 100000,
  "n_hvg": 0,
  "obs_value_filter": "tissue_general == 'lung' and is_primary_data == True",
  "organism": "Homo sapiens",
  "schema_version": "1.0",
  "seed": 0,
  "source": "cellxgene-census",
  "x_chunks": [
    1000,
    1000
  ],
  "x_compression": {
    "clevel": 3,
    "cname": "zstd",
    "codec": "blosc",
    "shuffle": "bitshuffle"
  },
  "shape": [
    100000,
    61497
  ],
  "obs_arrays": [
    "_index",
    "assay",
    "cell_type",
    "disease",
    "sex",
    "tissue"
  ],
  "var_arrays": [
    "_index",
    "feature_id",
    "feature_length",
    "feature_name",
    "feature_type",
    "n_measured_obs",
    "nnz",
    "soma_joinid"
  ]
}

Upload to HuggingFace Hub

This repo is intended for: KokosDev/single-cell-lung-zarr.

If you have a HuggingFace token locally, upload with:

python3.11 -m pip install huggingface_hub
HF_TOKEN=*** python3.11 build_lung_zarr.py --upload --repo-id "KokosDev/single-cell-lung-zarr"