nju-hefeng/singligent-catalog

Singligent — curated human single-cell RNA-seq metadata catalog

Harmonized, ontology-aligned, deduplicated metadata for 943,732 human scRNA-seq samples aggregated from 8 archives (GEO, EGA, NCBI, EBI BioStudies, CellxGene, PsychAD, HTAN, HCA), exposed at four record levels. Produced by Singligent (Nanjing University). Snapshot content fingerprint: f88b2025eda755b1 · built 2026-06-25.

• Portal: https://biobigdata.nju.edu.cn/singligent/
• Lab: https://compbio.nju.edu.cn/

⚠️ What this is — and is not

This bundle is the METADATA that describes what each dataset is (tissue, disease, cell type, donor, assay, provenance). It is NOT the raw sequencing data: count matrices / FASTQ / BAM remain at the source archives. The Singligent portal resolves exact, per-dataset download links (with sizes + checksums) to those originals.

Files

• unified_samples.parquet / unified_samples.csv.gz — Sample tier — cell-level metadata (one row per sample). (943,732 rows × 75 cols)
• unified_projects.parquet / unified_projects.csv.gz — Project tier — study-level groupings. (16,376 rows × 28 cols)
• unified_series.parquet / unified_series.csv.gz — Series tier — assay-level records with file pointers. (14,110 rows × 40 cols)
• unified_celltypes.parquet / unified_celltypes.csv.gz — Cell-type annotations — per-sample standardized (Cell Ontology) labels with counts. (378,029 rows × 8 cols)
• human_metadata.db — the complete relational SQLite snapshot (all tables + views).
• DATA_DICTIONARY.md — per-table column inventory.
• CHECKSUMS.sha256 — SHA-256 of every file.
• LICENSE.txt — Creative Commons Attribution 4.0 International (CC-BY-4.0).

Load

import pandas as pd
samples = pd.read_parquet("unified_samples.parquet")
# or CSV: pd.read_csv("unified_samples.csv.gz")
# or the full DB:  sqlite3.connect("human_metadata.db")

License & attribution

The compilation (harmonization, ontology alignment, dedup, schema) is released under Creative Commons Attribution 4.0 International (CC-BY-4.0) — please cite (see below). Records are derived from the source archives, which retain their own terms; cite the original studies when using specific datasets.

EGA: only controlled-access metadata is included here; access to EGA sequencing data requires Data Access Committee (DAC) approval at ega-archive.org.

Citation

He Feng. Singligent: a dual-agent portal for natural-language discovery of human single-cell RNA-seq metadata. Nanjing University, 2026. DOI: see the Zenodo record.

Reproducibility

All portal evaluations pin to snapshot fingerprint f88b2025eda755b1. Verify with: python scripts/export_catalog_release.py reports the same fingerprint, and CHECKSUMS.sha256 covers every file in this bundle.