scanpy

TAFFISH wrapper for Scanpy 1.12.2, the scverse Python toolkit for scalable single-cell gene-expression analysis.

Package Identity

• name: scanpy
• command: taf-scanpy
• kind: tool
• TAFFISH version: 1.12.2-r1
• container image: ghcr.io/taffish/scanpy:1.12.2-r1
• upstream: scverse/scanpy tag 1.12.2
• runtime: Python 3.12 with Scanpy 1.12.2
• app license: Apache-2.0
• upstream license: BSD-3-Clause

What Is Included

This image installs Scanpy from PyPI with the common optional extras used by standard offline workflows:

• scanpy[leiden] for igraph and leidenalg clustering support.
• scanpy[skmisc] for Seurat v3 highly-variable gene selection.
• Core Scanpy dependencies such as AnnData, NumPy, SciPy, pandas, scikit-learn, h5py, matplotlib, seaborn, statsmodels, UMAP, pynndescent, numba, and networkx.

The image does not bundle large reference atlases, model weights, notebooks, project datasets, or online services.

Usage

Show the TAFFISH package help:

taf-scanpy --help

Show the upstream Scanpy CLI help:

taf-scanpy -- --help
taf-scanpy scanpy --help

Run a Python one-liner inside the same container:

taf-scanpy python -c '"import importlib.metadata as md; print(md.version(\"scanpy\"))"'

Run a Scanpy analysis script:

taf-scanpy python analysis.py

The app keeps TAFFISH command mode enabled, so non-option-leading commands run inside the Scanpy container. This is useful for helper commands such as:

taf-scanpy python script.py
taf-scanpy scanpy --help
taf-scanpy pip list

For python -c, keep the code string quoted as a literal argument as shown above. For multi-line analysis, a script file or standard input is usually clearer.

Inputs And Outputs

Scanpy itself is a Python API, so inputs and outputs are determined by the script or upstream CLI command you run. Typical inputs include .h5ad, 10x Matrix Market directories, 10x HDF5 files, .loom, .mtx, .csv, .tsv, and other files supported by Scanpy/AnnData readers. Typical outputs include processed .h5ad files, tables such as marker-gene TSV/CSV files, and figures written by matplotlib/Scanpy plotting calls.

TAFFISH does not impose an output directory for this tool app. Your Python script controls where files are written; use explicit output paths when running inside workflows.

Minimal Example

taf-scanpy python - <<'PY'
import numpy as np
import pandas as pd
import scanpy as sc
from anndata import AnnData

adata = AnnData(np.array([[1, 0, 3], [0, 2, 1], [5, 0, 0]], dtype=float))
adata.obs_names = ["cell1", "cell2", "cell3"]
adata.var_names = ["geneA", "geneB", "geneC"]
adata.obs["group"] = pd.Categorical(["A", "A", "B"])

sc.pp.normalize_total(adata)
sc.pp.log1p(adata)
adata.write_h5ad("example.h5ad")
print(adata)
PY

Boundaries

r1 is an offline Scanpy runtime. It supports the core Python API and the upstream scanpy console script, including common preprocessing, AnnData I/O, PCA/neighbors, Leiden clustering, marker testing, HVG selection with skmisc, and static plotting through the headless matplotlib Agg backend.

The image does not include optional integration packages such as bbknn, harmonypy, magic-impute, scanorama, scrublet, Dask acceleration, RAPIDS/GPU dependencies, JupyterLab, legacy louvain, or single-cell reference datasets. Those can be added in future specialized apps or flows if they become part of a stable TAFFISH workflow. Louvain is intentionally not bundled in r1 because its current PyPI path needs source-build tooling on Python 3.12; Leiden is the packaged clustering path.

Network access is not required for the packaged smoke path. User scripts may use the network only if the container backend and site policy allow it; online downloads are not part of the reproducible r1 contract.

Citation

Wolf FA, Angerer P, Theis FJ. SCANPY: large-scale single-cell gene expression data analysis. Genome Biology 19, 15 (2018). doi: 10.1186/s13059-017-1382-0; PMID: 29409532.