---
pretty_name: SampleExplorer databases (ARCHS4 v2.7, release 1)
language:
  - en
tags:
  - RNA-seq
  - GEO
  - ARCHS4
  - gene-set-search
  - transcriptomics
  - SampleExplorer
size_categories:
  - 1M<n<10M
---

# SampleExplorer databases (ARCHS4 v2.7, release 1)

Databases for [SampleExplorer](https://github.com/wlchin/SampleExplorer): one SQLite file per species.
Each is a single file with nothing to unzip. This repository is the home of the files: every release is a tag
here, and the desktop app's catalogue points at these tags. Check a download against the sha256 below.

| File | Species | Studies | Samples | Size | sha256 |
|---|---|---|---|---|---|
| `sampleexplorer-human-v2.7-r1.sqlite` | human | 38,761 | 1,098,771 | 1,556,414,464 bytes (1.6 GB) | `ba3c7446968d719c965241dd3dcc16a040257976c5037e7026d345e7fd3f488c` |
| `sampleexplorer-mouse-v2.7-r1.sqlite` | mouse | 40,620 | 1,225,044 | 1,717,370,880 bytes (1.7 GB) | `852056fff702f0cad8e39ccd8d81814a81d6529d30ab12647886a3e35b14f530` |

## Download

With the Hugging Face CLI (resumable, checks the hash for you):

```
uvx --from huggingface_hub hf download Wlchin/sampleexplorer-databases sampleexplorer-human-v2.7-r1.sqlite --repo-type dataset --local-dir .
uvx --from huggingface_hub hf download Wlchin/sampleexplorer-databases sampleexplorer-mouse-v2.7-r1.sqlite --repo-type dataset --local-dir .
```

From Python:

```python
from huggingface_hub import hf_hub_download
path = hf_hub_download("Wlchin/sampleexplorer-databases", "sampleexplorer-human-v2.7-r1.sqlite",
                       repo_type="dataset", revision="r1")
```

With curl (resumes with `-C -`):

```
curl -L -C - -o sampleexplorer-human-v2.7-r1.sqlite https://huggingface.co/datasets/Wlchin/sampleexplorer-databases/resolve/r1/sampleexplorer-human-v2.7-r1.sqlite
```

Check a download with `shasum -a 256 <file>` on macOS and Linux, or `certutil -hashfile <file> SHA256` on Windows.

## Use

Open a file in the SampleExplorer desktop app with **Locate database file…**, or read it with any `sqlite3`.
Open both files to search across species.

## What each file holds

- GEO study and sample text
- the per-sample gene index for gene-set search
- MiniLM text vectors and the MiniLM model
- the response bases for the Same response search
- the BM25 keyword index
- MedCPT article vectors and the MedCPT query encoder
- the 50 MSigDB Hallmark pathway sets and the pathway bank behind Compare pathways

The file's `meta` table records the ARCHS4 version, the build rules, content hashes and the sha256 of every bundled model
and of the pathway sources.

## Source data

Expression and sample metadata come from [ARCHS4](https://maayanlab.cloud/archs4/) v2.7 (Lachmann et al., 2018,
*Nature Communications* 9:1366), which reprocesses public RNA-seq from GEO and SRA. Study and sample text come from
[GEO](https://www.ncbi.nlm.nih.gov/geo/). Please cite ARCHS4 and GEO alongside SampleExplorer when you use these files.

## Pathway gene sets: sources and attribution

The pathway data (the `pathway_sets` table, the `pathway_bank` array and meta `pathways`) come from two sources:

- **MSigDB Hallmark v2023.2.Hs** (`h.all.v2023.2.Hs.symbols.gmt`, 50 sets) from the
  [Molecular Signatures Database](https://www.gsea-msigdb.org/gsea/msigdb), UC San Diego and Broad Institute, under
  [CC BY 4.0](https://creativecommons.org/licenses/by/4.0/). Each set is kept as its genes in the file's gene universe,
  and the file stores directions computed from them; the set names are unchanged. Cite Liberzon et al., 2015,
  *Cell Systems* 1:417–425 ([doi:10.1016/j.cels.2015.12.004](https://doi.org/10.1016/j.cels.2015.12.004)) and MSigDB
  (Subramanian et al., 2005, *PNAS* 102:15545–15550).
- **MGI's mouse–human homology report** (`HOM_MouseHumanSequence.rpt`, dated 2026-10-28), mouse file only: it maps the
  human sets to one-to-one mouse orthologues. From [Mouse Genome Informatics](https://www.informatics.jax.org), The Jackson
  Laboratory; cite Baldarelli et al., 2024, *Genetics* 227:iyae031
  ([doi:10.1093/genetics/iyae031](https://doi.org/10.1093/genetics/iyae031)).

The full notice, with the changes made, the sha256 of both source files and the MGI data citation, is in
`LICENSES/CC-BY-4.0-MSigDB-Hallmark.txt`.

## Bundled models and their licences

Two text encoders are stored inside each file as ONNX exports, so the app needs no PyTorch and no separate download.
Converting them to ONNX and bundling them is the only change made to them.

| Model | Used for | Source | Licence |
|---|---|---|---|
| paraphrase-MiniLM-L6-v2 | search by meaning | [sentence-transformers/paraphrase-MiniLM-L6-v2](https://huggingface.co/sentence-transformers/paraphrase-MiniLM-L6-v2), revision `c9a2bfebc254878aee8c3aca9e6844d5bbb102d1` (Reimers & Gurevych, UKP Lab) | Apache-2.0 |
| MiniLM base weights | base of the above | Microsoft MiniLM | MIT |
| MedCPT query encoder | keyword + semantic search | [ncbi/MedCPT-Query-Encoder](https://huggingface.co/ncbi/MedCPT-Query-Encoder) (Jin et al., 2023) | public domain (US Government work) |
| MedCPT article encoder | article vectors (vectors stored, encoder not bundled) | [ncbi/MedCPT-Article-Encoder](https://huggingface.co/ncbi/MedCPT-Article-Encoder) | public domain (US Government work) |

The Apache-2.0 licence text is in `LICENSES/Apache-2.0.txt` and the MIT text in `LICENSES/MIT-MiniLM.txt` in this
repository, as the Apache licence requires for redistribution.

## Versions

Each release is a tag (`r1`, `r2`, …) on this repository. A released file is never changed: a new build gets a new file
name (`-r2`), a new commit and a new tag, and the old tag keeps serving the old bytes. Download by tag or by commit
hash, not from `main`, so a link always returns the same file. Every push is a commit, so the full history of the
repository is the release history.

## Citation

Until the paper is out, cite this repository by its DOI (generated from the repository settings; added here once minted)
and the release tag:

> SampleExplorer databases (ARCHS4 v2.7, release 1), tag `r1`. Hugging Face. DOI: to be added.

How to cite the paper: added when the paper is published.