Tutorials

Tutorials#

HTML on Read the Docs is pre-executed (tables and figures already there). Re-run locally only if you have the .h5ad files at the repo root.

Pick a notebook#

If you want…	Open
No `spliced`/`unspliced` layers yet — velocyto, kb-python, STARsolo, alevin-fry, or labeling data	Preparing spliced/unspliced data
Full DE → mechanism story (human LPS-PBMC)	Partition DE by mechanism — transcription vs stabilization
Same API with real DE hits (mouse, 3 vs 3)	Partition by mechanism — GA vs Ctrl
Underpowered design (empty DE list on purpose)	Partition by mechanism — when DE finds nothing
DE + enrichment, no nascent layers	Differential expression + enrichment, no velocity layers
Plot gallery only	Visualization gallery
Gene overlap across DE backends	Gene overlap across DE methods

If you are new: read Quickstart, then either Partition DE by mechanism — transcription vs stabilization or Partition by mechanism — GA vs Ctrl. Open the SCI/EC notebook only if you want the empty-list lesson. If your AnnData does not have nascent layers at all, start with Preparing spliced/unspliced data instead.

SCI (EC) and GA share the same entry point (partition_de_by_mechanism). Whether genes are selected depends on the design, not on the function name.

Partition notebooks#

Notebook	Design	What to expect
LPS-PBMC	Human 10x, resting vs LPS 4 h	Full path: DE, labels, programs, absolute placement
GA vs Ctrl	Mouse, 3 individuals per group	Real DE hits, programs, enrichment
SCI vs UN (EC)	Mouse endothelium, 3 vs 3	Capture OK but 0 DE genes; sample- vs cell-level contrast

Run locally#

pip install "scatrans[pseudobulk,gsea]"
# or from a clone: pip install -e ".[dev,pseudobulk,gsea]"
jupyter lab docs/tutorials/

Put data files at the repository root. Notebooks load them with a relative path, for example sc.read_h5ad("../../EC.h5ad").

File	Used by
`GSE226488_PBMC_tutorial_subset.h5ad`	LPS-PBMC
`GA_test.h5ad`	GA
`EC.h5ad`	SCI partition, gene UpSet
`kang_ifnb_tutorial_subset.h5ad`	Standalone DE + enrichment

These files are large and are not on PyPI. Citations: References & Data Sources. Without them you can still read the HTML on Read the Docs. The visualization gallery does not need an external file.

Rough runtime with data on disk: LPS / GA about 10–30 min (calibration is slower); EC / UpSet about 2–15 min; standalone DE + enrichment about 5–15 min; synthetic plots about 1–3 min.