AIVIN-UCSD/liver-references

HCC Multi-Source Database Integration

Curated, cross-referenced index of hepatocellular carcinoma (HCC / liver cancer) datasets harvested from many public repositories, plus the raw single-cell downloads that were pulled for the project.

Access note. This is a private dataset repository. Some source catalogs point to controlled-access studies (e.g. EGA, ICGC LIRI-JP). Only metadata / catalog tables for those sources are stored here — no controlled raw genomic files. Do not make this repo public or redistribute controlled-access raw data without the originating DAC's approval.

Contents

Layer	What it is	Location
Master index	Unified table of every dataset (47 columns, ~11.4k rows)	HCC_Database_Master_*.xlsx (versioned), latest = *_T16_diseasefix.xlsx
Source catalogs	Per-repository metadata tables + a README.md each	ARCHS4/, DepMap/, EGA/, ENA/, ICGC_LIRI_JP/, LINCS_CMap/, MSigDB/, ChineseCohorts/, SingleCellPortal_SCP/, UKB_Schemas/, JJBC/
Build outputs	Intermediate crawl / dedup / validation artifacts	outputs/
Raw single-cell	Downloaded expression matrices (.h5ad, .mtx, .rdata, .gz, …)	SingleCellAtlasDownload/, SingleCellExpressionAtlas/, SingleCellPortal/, LiverCellAtlasDownload/
Raw bulk / omics	Sequencing & proteomics downloads (.fastq.gz, .wiff/.scan, archives)	GSADownload/, KDA/, PRIDEDownload/
Build scripts	The Python crawlers/mergers used to assemble the DB	*.py in the root

Data volume: ~291 GB across ~15k files after excluding partial-download junk (*.tmp/*.part, ~10 GB). Largest single file ≈ 10 GB (< HF's 50 GB/file limit).

Master table schema (HCC_Database_Master)

Single sheet named Master. Key columns:

source_db, source_file, accession, granularity, tier, tier_confidence, title, keywords, description, organism, disease, tissue, sample_type, technology, platform, n_samples, cell_count, n_files, size_MB, release_date, doi, pmid, bioproject, geo, sra, http_link, ftp_link, page_url, access_status, dup_group_id, dup_role, disease_inferred, … (47 cols total)

• tier — relevance ranking (curator-assigned), with tier_confidence / tier_basis.
• dup_group_id / dup_role — cross-source de-duplication grouping.
• accession — the canonical repository ID (GEO/SRA/EGA/…); join key across catalogs.

Versioning

Master snapshots are suffixed _T{n}_{step} in build order (T1_modified … T16_diseasefix). The highest T-number is the current release; earlier ones are kept for provenance.

How this was built

See the root-level *.py scripts (one per source: fetch_hcc_geo.py, ega_crawl_hcc.py, ena_crawl_hcc.py, scp_crawl_hcc.py, msigdb_filter_hcc.py, deep_dedup.py, …) and each source folder's own README.md.

Citation / provenance

Each record carries its own doi / pmid / accession. Cite the original study, not this index. This repository is an aggregation aid, not an original data source.