Cellucid Web Viewer

WebGL-based 3D visualization for single-cell data. Explore UMAP embeddings with interactive coloring, filtering, gene expression overlays, and KNN connectivity graphs.

Live demo: cellucid.com

Features

• Real-time rendering of millions of cells with adaptive LOD
• Gene expression overlays with efficient sparse matrix handling
• Categorical and continuous cell metadata coloring
• Interactive filtering and cell selection
• KNN connectivity edge visualization
• Multi-dimensional support (1D timelines, 2D, 3D)
• Works in browser without Python (file picker) or with Python (Jupyter, CLI)

Quick Start

Option 1: Browser File Picker (No Setup)

1. Go to cellucid.com
2. Click "Browse local data..."
3. Select a pre-exported folder, .h5ad file, or .zarr store

Option 2: Python CLI

pip install cellucid

# Serve h5ad directly
cellucid-anndata /path/to/data.h5ad

# Serve zarr directly
cellucid-anndata /path/to/data.zarr

# Or serve pre-exported data
cellucid serve ./my_export

Option 3: Jupyter Notebook

from cellucid import show_anndata
show_anndata(adata)  # In-memory or file path

All 14 Loading Options

Cellucid supports 6 deployment modes, each with support for pre-exported binary data, h5ad files, and zarr stores:

#	Method	Exported	h5ad	zarr	Python	Lazy Load	Performance
1	Local Demo (GitHub)	✅	-	-	No*	Yes	Best
2	Remote Demo (GitHub)	✅	-	-	No*	Yes	Best
3	Browser File Picker	✅	-	-	No	Yes	Best
4	Browser File Picker	-	✅	-	No	No	Slower
5	Browser File Picker	-	-	✅	No	No	Slower
6	Server CLI	✅	-	-	Yes	Yes	Best
7	Server CLI	-	✅	✅	Yes	Yes	Good
8	Python serve()	✅	-	-	Yes	Yes	Best
9	Python serve_anndata()	-	✅	✅	Yes	Yes	Good
10	Jupyter show()	✅	-	-	Yes	Yes	Best
11	Jupyter show_anndata()	-	✅	✅	Yes	Yes	Good

* Python required for initial export, not for viewing

Summary by method:

Method	Exported	h5ad	zarr	Total
Local/Remote Demo	✅	-	-	2
Browser File Picker	✅	✅	✅	3
Server CLI	✅	✅	✅	3
Python serve	✅	✅	✅	3
Jupyter	✅	✅	✅	3
Total				14

Key Notes

• Browser h5ad/zarr (Options 4-5): Entire file loaded into browser memory - no lazy loading possible due to JavaScript limitations. Best for datasets < 100k cells.
• Python h5ad/zarr modes (Options 7, 9, 11): True lazy loading via AnnData backed mode (h5ad) or zarr's native chunked access. Recommended for large datasets.
• Pre-exported data: Always fastest - recommended for production and sharing.
• zarr stores: Can be a directory (.zarr) or a file - the Python server auto-detects the format.

h5ad / zarr Requirements

• Required: obsm['X_umap'] or obsm['X_umap_3d'] (shape: n_cells × 2 or 3)
• Optional: obs (cell metadata), X (expression matrix), obsp['connectivities'] (KNN graph)

Saving as zarr

# Save AnnData as zarr store
adata.write_zarr("data.zarr")

Pre-export for Performance

For best performance, especially with large datasets:

from cellucid import prepare
prepare(adata, output_dir="./my_export", compress=True)

Repository Structure

cellucid/
├── index.html              # Single-page app
├── assets/
│   ├── js/
│   │   ├── app/            # UI, state management
│   │   ├── data/           # Data loaders (binary, h5ad)
│   │   └── rendering/      # WebGL renderer
│   └── exports/            # Sample datasets

Development

python -m http.server 8000
open http://localhost:8000

Python Package

See theislab/cellucid-python for the companion Python package.

License

Proprietary - All rights reserved