scATrans

scATrans answers a simple follow-up after differential expression (DE): among the genes that changed, which look transcription-driven and which look stabilization-driven? It uses the nascent (unspliced) residual on top of a normal DE step—something total-count fold change alone cannot sort out.

Pathway- or program-level summaries (gene_sets=) are usually more useful than single-gene labels. Without spliced/unspliced layers, the package still runs ordinary DE, enrichment, and plots.

Docs: Read the Docs.

What you need

• Python 3.9+ (tested 3.9–3.12)
• An AnnData object with a condition column in .obs
• For mechanism analysis: spliced/unspliced (or mature/nascent) layers
• Nothing yet? scat.datasets.load_toy() ships a synthetic example — see First run.

Install

pip install scatrans
# or: conda install -c conda-forge -c bioconda scatrans

Optional extras (PyDESeq2, GSEA, scVelo, …): installation guide.

First run

This uses a bundled synthetic AnnData (load_toy()). No download. Swap in your own object after the call works.

import scatrans as scat

adata = scat.datasets.load_toy()  # Control/Disease, sample, spliced/unspliced

result = scat.partition_de_by_mechanism(
    adata,
    groupby="condition",
    target_group="Disease",
    reference_group="Control",
    organism="mouse",          # "human" for human gene symbols
    sample_col="sample",       # toy has this column; pass yours when you have replicates
)
print(result.regime)           # capture quality (reliability in [0, 1])
print(result.selected.head())  # DE genes + mechanism labels
print(result.summary())        # cutoffs used, how many genes were selected

What to expect:

• result.selected is the DE gene list. Mechanism columns annotate it; they do not drop genes.
• Default gates are padj < 0.05 and logFC > 1.0. If selected is empty, lower logfc_cutoff (for example 0.25) or check the contrast.
• Omit sample_col only when you have no biological replicates (cell-level DE).
• Per-gene mechanism_class is exploratory. For claims, pass gene_sets= and/or run program_mechanism_permutation_calibrated.

No spliced/unspliced (or mature/nascent) layers? You can still run DE and enrichment — see Quickstart. To build those layers: Preparing spliced/unspliced data.

Next: Quickstart · Tutorials · FAQ

Status

0.10.x (Beta). Import as import scatrans as scat and stick to names in scatrans.__all__, scat.pl, and scat.qc. Details: API stability.

Citation

Please cite the preprint:

Li, Z., James, A. W. & Li, S. scATrans: annotating single-cell differential expression as transcription- or stabilization-weighted using unspliced RNA. bioRxiv (2026). doi:10.64898/2026.08.03.740741

For the software itself, cite the Zenodo DOI above. Pin the installed version in Methods (this tree is scatrans==0.10.14). See CITATION.cff.

License

Software: Apache License 2.0. Bundled GO/KEGG data may carry separate terms—see the license page before commercial redistribution.

Author

Zhao Li (李钊)
Email: leelieber@gmail.com