Installation

Installation#

Requirements#

  • Python 3.9+ (CI covers 3.9–3.12)

  • A normal scientific stack (NumPy, SciPy, pandas, AnnData, scanpy) — pulled in as dependencies

Install#

PyPI (recommended):

pip install scatrans

Bioconda:

conda install -c conda-forge -c bioconda scatrans

Conda installs the core package. Optional extras below are PyPI extras; add them with pip or install the matching conda packages yourself.

Optional extras#

Install only what you need:

pip install "scatrans[pseudobulk]"      # PyDESeq2 (replicate-aware DE)
pip install "scatrans[gsea]"           # GSEA via gseapy
pip install "scatrans[memento]"        # Memento DE backend
pip install "scatrans[advanced]"       # scVelo (mode="advanced")
pip install "scatrans[gene_features]"  # gtfparse for custom GTF tables

Combine tags as needed, e.g. "scatrans[pseudobulk,gsea]".

Bundled mouse/human gene-feature tables support optional length/intron bias correction. Custom GTF tables: Gene Feature Attachment and CLI.

Check the install#

import scatrans as scat

print(scat.__version__)
adata = scat.datasets.load_toy()
result = scat.partition_de_by_mechanism(
    adata,
    groupby="condition",
    target_group="Disease",
    reference_group="Control",
    organism="mouse",
    sample_col="sample",
)
print(result.summary())

That should print a small selected-gene count without downloading anything.

After install#

Step

Page

First analysis

Quickstart

Which function / backend?

FAQ / Troubleshooting

Real-data notebooks

Tutorials

Full workflow knobs

User Guide

Development install#

git clone https://github.com/leelieber2025/scATrans.git
cd scATrans
pip install -e ".[dev]"

Versioning#

The single source of truth is src/scatrans/_version.py. Runtime scatrans.__version__, packaging metadata, and docs release strings all read it. For a release: bump __version__, update CHANGELOG.md, then python -m build or python scripts/make_release_zips.py.

Logging#

import logging
logging.getLogger("scatrans").setLevel(logging.INFO)

Quick data check (before mechanism analysis)#

import scatrans as scat

print(scat.qc.unspliced_global(adata))
r = scat.qc.regime_diagnosis(adata)
print(r["regime"], r["reliability"], r["message"])
# regime: "ok" | "low_unspliced" | "high_unspliced"

partition_de_by_mechanism always runs this check and stores it as result.regime. Low reliability does not stop the run; it down-weights mechanism confidence.