Standard Pancreas sample dataset#

Cellucid’s public sample catalog includes the pancreas dataset used by the scVelo pancreas vignette. It is a complete known-good example for dimensional navigation, categorical and continuous metadata, gene coloring, connectivity, and RNA-velocity rendering. The catalog id is exactly pancreas; Suo remains the catalog default. Its published identity contains exactly 3,696 cells and 3,753 genes.

Load it#

  1. Open an ordinary Cellucid startup and dismiss the welcome overlay.

  2. Open Session.

  3. Under Sample datasets, select Pancreatic endocrinogenesis (scVelo).

  4. Wait until the loaded-dataset name changes to Pancreatic endocrinogenesis (scVelo). The compact Session statistics round both counts to 4K; Coloring & Filtering reports the exact Showing all 3,696 points, while the published identity and provenance pin the exact inventory of 3,753 genes.

The sample opens in its declared default 3D embedding with Orbit navigation. Selecting 1D or 2D changes the dimension-specific default to Planar, unless you explicitly chose another navigation mode. No camera path is created or played automatically.

Verify the scientific controls#

The public dataset exposes:

  • source-owned categorical observations clusters_coarse (5 categories) and clusters (8 categories);

  • the derived categorical convenience alias cell_type (8 categories), whose category order and per-cell codes exactly equal source-owned clusters;

  • source-owned continuous observations S_score and G2M_score;

  • 3,753 ordered, source-marked, nonconstant genes for on-demand coloring;

  • 33,476 exact weighted connectivity edges;

  • independently declared 1D, 2D, and 3D embeddings; and

  • one dimension-matched vector field, velocity_umap, in 1D, 2D, and 3D.

To inspect velocity, open Visualization, keep Render mode set to Points, find Vector Field Overlay, and enable Show overlay. The field selector reports velocity_umap. The overlay is a qualitative view of the projected stochastic velocity; it does not replace scVelo’s numerical diagnostics or uncertainty checks.

Cellucid showing the Pancreatic endocrinogenesis (scVelo) sample in its 3D Orbit view, colored by clusters, with the velocity_umap overlay enabled and vector controls visible.

The standard Pancreas sample in Build 2026-07-27.1: clusters supplies eight categories, Show overlay is enabled, velocity_umap is selected, and the dimension-owned controls report 3D with Orbit navigation.#

What is source-owned and what is derived#

The source is the commit-pinned E15.5 H5AD from the scVelo notebooks:

https://raw.githubusercontent.com/theislab/scvelo_notebooks/f6cad69fd509be44c8453205309f4c3c3c37ba34/data/Pancreas/endocrinogenesis_day15.h5ad

Its exact 2D UMAP, 50-dimensional PCA, cell and gene order, four source-owned observation fields (clusters_coarse, clusters, S_score, and G2M_score), highly-variable-gene marker, expression values, and weighted connectivity graph are retained as source-owned inputs. The public cell_type field is a derived convenience alias of clusters, with exactly the same category order and per-cell codes; it is not a fifth source-owned field. The 1D and 3D UMAPs are independently optimized deterministic embeddings; neither is sliced, padded, or copied from the 2D coordinates. Velocity is computed on a separate analysis-only working copy and projected independently onto each exact embedding. It never replaces the public expression inventory.

The source contains 27,998 ordered genes. The network payload intentionally ships the exact 3,753 columns that are both marked highly_variable_genes == "True" and nonconstant. This removes unmarked and constant per-gene requests without changing cell order, selected-gene order, or the selected expression values.

Provenance and reproducibility#

The canonical provenance is sources/pancreas.json in cellucid-datasets. It pins:

  • the source commit, URL, SHA-256, byte size, and biological citation;

  • the exact Python and scientific-package environment;

  • deterministic embedding and velocity parameters;

  • semantic digests for coordinates, vectors, metadata, genes, expression, and connectivity;

  • the exact Cellucid 0.9.1 producer-source digest; and

  • the complete 3,774-file, 2,477,205-byte generation digest 268a36b0c62f1e872bc4ebc271283fe09b9036c2b8db2b7bbf2d7d737556c865.

The source data correspond to GSE132188. For scientific use, cite Bastidas-Ponce et al. (2019) in addition to the tools used in your analysis.

Troubleshooting#

  • If the sample is not listed, the configured exports host is not serving the current datasets.json; include the footer Build value in a bug report.

  • If the overlay controls are absent, confirm that the loaded dataset name is Pancreatic endocrinogenesis (scVelo) and that Points is the active render mode.

  • If Show overlay disables after a dimension change, wait for that dimension’s embedding and velocity_umap payload to finish loading before retrying.

  • If points or flow look biologically surprising, compare the exact dimension with the scVelo analysis and inspect velocity confidence before interpreting directionality.

For the complete overlay control reference, see Vector Field / Velocity Overlay (GPU Particle Overlay). For general sample and remote loading failures, see Troubleshooting (data loading).