Ctrlk
For the complete documentation index, see llms.txt. This page is also available as Markdown.

Celldega

An open-source visualization toolkit and spatial biology exploration engine developed by the Broad Institute of MIT and Harvard (Fernandez, Ishar, Wang, Ben Saad, Lipinski & Farhi, bioRxiv 2026). Celldega is engineered to address the scalability bottleneck of client-side spatial omics and multiplexed digital pathology datasets (>1 billion transcripts, millions of segmented cells) by leveraging browser-native streaming, GPU-accelerated rendering, and specialized columnar storage formats.

Architectural Highlights & Innovations

1. DegaFiles and Row Group Spatial Indexing

Traditional spatial transcriptomics data storage splits multiplexed datasets into tens of thousands of fragmented files or bulky monolithic tables that overload browser memory. Celldega introduces DegaFiles, an open storage convention built on Apache Parquet and GeoParquet:

  • Consolidates 50,000+ individual coordinate/feature files into ~10 structured Parquet tables.

  • Employs spatial indexing where spatial tile coordinates map directly to Parquet row groups (row_group_index = tile_x * num_tiles_y + tile_y).

  • Utilizes byte-range HTTP streaming via WebAssembly (parquet-wasm and Apache Arrow), fetching only the precise byte chunks needed for the user's active viewport without requiring server-side Python kernels.

2. GPU-Accelerated Multi-Layer Rendering

Built on deck.gl and HTML5 Canvas, Celldega renders hundreds of millions of spatial data points, polygonal cell segmentations, and multi-channel image pyramids concurrently at 60 fps:

  • Spatial Alignment: Co-registers high-resolution whole-slide brightfield/fluorescent imagery with single-cell transcriptomic annotations.

  • Dynamic Shading & Gating: Real-time client-side expression thresholds and cluster re-coloring without server re-computation.

3. Specialized Analytical Views

Celldega provides distinct visualization modalities:

  • Landscape View: 2D multi-layer spatial exploration of tissue sections, combining cell segmentation masks and individual transcript dots.

  • Yearbook View: High-throughput single-cell gallery displaying isolated morphology across phenotypic clusters.

  • CellCloud & NeighborhoodCloud: 3D orbital embedding visualizers linking dimensionality-reduced manifolds (UMAP/t-SNE) directly back to spatial coordinates.

  • Clustergram & Composition: Interactive heatmaps displaying cell type compositions across histological tissue zones.

  • Enrich: Direct integration with Enrichr for immediate pathway and gene-set enrichment queries from selected spatial regions.

Comparison with Existing Spatial Systems

Dimension
Cytario
Celldega
Vitessce

Primary Scope

Image Management System (IMS) & clinical WSI viewer

Spatial omics & multiplexed transcriptomics analytics

General bioimaging & spatial omics dashboard

Backend Storage

S3 object store (OME-Zarr, OME-TIFF, GeoTIFF)

DegaFiles (GeoParquet / Parquet with row-group indexing)

Zarr / AnnData / OME-TIFF

Streaming Mechanism

DuckDB-WASM & Viv multi-scale tiles

Parquet-WASM byte-range streaming via deck.gl

Viv multi-resolution loaders

Target Workload

Petabyte-scale clinical digital pathology workflows

Billion-transcript spatial exploratory analysis

Multimodal exploratory dashboards

PreviousM-Optimus
NextHistoCAM

Last updated