AnnZarro

A read-only browser viewer for AnnData in zarr (or h5ad) built around the matrices other viewers leave out: cell x cell kernels and distances (obsp) and gene x gene similarities (varp), explored one focused cell or gene at a time and linked to the cell x gene layers.

Documentation: annzarro.readthedocs.io (tutorials, user guide, data preparation, deployment and reference). Developed by the Setty Lab.

Install

pip install annzarro              # or 'annzarro[remote]' for s3://, gs:// and https:// stores

This installs AnnZarro on your own computer or on a server. Or download the desktop app, or set up a shared server, below.

From source: git clone https://github.com/settylab/annzarro.git && pip install -e ./annzarro. Details: installation, desktop app.

Desktop app or server

Desktop app (download the latest release):

macOS: Apple silicon, Intel, or brew install --cask settylab/annzarro/annzarro. Linux: AppImage, .deb. All files and checksums.

• Standalone on your laptop, no Python needed
• Works offline
• Your data stay on your computer
• No login, no set-up
• Opens local .zarr stores

Server (one server, many users in the browser; install):

• Multi-user access in the browser, with login
• Panel sets shared and exchanged between users
• Large datasets stay on the server; only the slices a view needs travel to the browser
• Per-user dataset access control and an admin page
• Less memory on each user's computer

What it does

• Colours any plot by the focused cell's row of an obsp matrix or the focused gene's row of a varp matrix.
• Plots any obs, var, obsm, varm column or layers row/column against any other, spatial coordinates included.
• Links cells and genes: one cell's values across genes, one gene's values across cells, in every layer.
• Filters cell and gene tables with AND/OR conditions that mask the plots; saves layouts as panel sets and share links.
• Reads only the chunks behind what is on screen, so cost follows the view, not the dataset size; locally or from S3, GCS and HTTP.
• Never writes to your data and never runs code: a lab server can show datasets without giving write or compute access.

Quickstart

mkdir -p ~/annzarro-data
ln -s /path/to/your.zarr ~/annzarro-data/
annzarro start --data-dir ~/annzarro-data      # opens http://127.0.0.1:8000

Walkthrough with the demonstration data: quickstart.

Prepare your data

Any AnnData written with adata.write_zarr(...) opens as is. Which slot feeds which view, how to precompute obsp/varp matrices and how to chunk for speed: preparing data.

Deployment

• Desktop app: one person, data on the same computer. Set up
• Personal server: annzarro start on a laptop or an HPC node, reached over an SSH tunnel. Set up
• Lab server: gunicorn behind HTTPS, read-only, with login and shared panel sets. Set up

Documentation

annzarro.readthedocs.io: tutorials on cell and gene similarity, the user guide, data preparation, deployment, and the CLI, configuration and HTTP API reference.

Citing AnnZarro

If you use AnnZarro, please cite the preprint:

Otto, D. J., Baasri, S. and Setty, M. AnnZarro: scalable, interactive exploration of cell-by-cell and gene-by-gene relationships in single-cell data. bioRxiv (2026). doi:10.64898/2026.10.10.758158

@article{otto2026annzarro,
  author  = {Otto, Dominik J. and Baasri, Siddharth and Setty, Manu},
  title   = {{AnnZarro}: scalable, interactive exploration of cell-by-cell and gene-by-gene relationships in single-cell data},
  journal = {bioRxiv},
  year    = {2026},
  doi     = {10.64898/2026.10.10.758158},
  url     = {https://doi.org/10.64898/2026.10.10.758158}
}

To cite the software itself, for example the exact release you ran, use its Zenodo DOI. Each release is archived on Zenodo with its own version DOI (v0.5.1: 10.5281/zenodo.23386903); the concept DOI below always resolves to the latest release.

@software{otto_annzarro_software,
  author    = {Otto, Dominik J. and Baasri, Siddharth and Setty, Manu},
  title     = {AnnZarro},
  publisher = {Zenodo},
  doi       = {10.5281/zenodo.23222216},
  url       = {https://doi.org/10.5281/zenodo.23222216}
}

Citation metadata are also in CITATION.cff (GitHub's "Cite this repository"), which names the preprint as the preferred citation.

Development

npm ci once (Node.js 22+), then python -m pytest runs the Python tests, every JS suite and ESLint. See running the tests.

Licence

MIT (LICENSE). The web interface bundles unmodified third-party libraries under their own permissive licences, listed in annzarro/THIRD_PARTY_LICENSES/.