CytePool

Test which genes, cell types and pathways change between conditions in single-cell and bulk RNA-seq.

Cells from one donor are not independent replicates. CytePool pools each sample and cell type into a cube of pseudobulk profiles, and runs limma, edgeR, DESeq2, dream, variancePartition, propeller, fgsea, GSVA, decoupler, metafor and SumRank in Python on numpy arrays.

Install

Requires Python 3.13 or newer:

pip install cytepool

For AnnData, MuData and plotting, install pip install "cytepool[mudata,plot]". See Installation.

Example

Pool cells into a cube, then test a stimulation across cell types with limma-voom, pairing samples by donor:

from cytepool import Design, camera, from_cells, limma_voom, read_gmt

cube = from_cells(
    counts,  # cells x genes, e.g. adata.X
    sample_labels=cell_sample,  # e.g. adata.obs["sample"]
    group_labels=cell_type,  # e.g. adata.obs["cell_type"]
    samples=samples,  # sample_id, donor, condition
    features=genes,  # feature_id and symbol, or adata.var
)
design = Design(
    cube.samples,
    "~ donor + condition",
    levels={"condition": ["ctrl", "stim"]},
    contrasts={"stim_vs_ctrl": "conditionstim"},
)
result = limma_voom(cube, design)
table = result.table("Monocytes", "stim_vs_ctrl")  # topTable columns

# Test pathway response in monocytes from a GMT file.
hallmark = read_gmt("h.all.v2026.2.Hs.symbols.gmt")
pathways = camera(result, hallmark, group="Monocytes", id_column="symbol")

If you already pooled profiles in scanpy, decoupler or LIANA, import them with from_anndata or from_mudata. See the Quick start and Gene sets and pathways.

What CytePool answers

Question	Start with	Other CytePool functions	R and Python equivalents
Pool cells or import profiles	from_cells, from_bulk	from_anndata, from_mudata, write_cube, read_cube	scanpy get.aggregate, decoupler pp.pseudobulk, LIANA adata_to_views
Which genes change in each cell type?	limma_voom	limma_trend, filter_by_expression, tmm_factors	limma voom, lmFit, eBayes, topTable; edgeR filterByExpr, calcNormFactors
Which genes change in bulk RNA-seq?	limma_voom on from_bulk	fit_quasi_likelihood, quasi_likelihood_f_test, fit_deseq2, deseq2_wald_test, deseq2_likelihood_ratio_test, treat	edgeR glmQLFit, glmQLFTest; DESeq2 DESeq, results; limma treat
Which genes change, with several samples of each donor?	dream_voom	fit_mixed_model, fixed_effect_tests, fixed_effect_f_test	variancePartition dream, voomWithDreamWeights; lme4 lmer, lmerTest, pbkrtest
What drives the variation between samples?	variance_partitioning	principal_components, plot_mds, plot_pca	variancePartition fitExtractVarPartModel; DESeq2 plotPCA; limma plotMDS
Which cell types expand or contract?	propeller	plot_composition	speckle propeller
Do the genes of a pathway respond to a contrast?	camera	fry, roast, mroast, romer, over_representation	limma camera, fry, roast, mroast, romer, kegga
Do the genes of a pathway gather along a ranking?	gsea	plot_enrichment, plot_gsea_table	fgsea fgseaMultilevel, plotEnrichment, plotGseaTable
How active is a pathway in each sample?	gsva	ssgsea, combined_z_scores, plage, limma	GSVA gsva, ssgseaParam, zscoreParam, plageParam
Which regulators and pathways are active?	ulm	mlm, weighted_aggregate, collectri, progeny, limma	decoupler mt.ulm, mt.mlm, mt.waggr; CollecTRI, PROGENy
What do several studies say together?	meta_analysis	CubeCollection, analyse_each_study, combine_p_values	metafor rma.uni; metap
Which changes replicate across studies?	sumrank	replication, partial_conjunction, proportion_of_true_nulls, held_out_prediction	SumRank; qvalue pi0est

Choosing a method compares the alternatives for each question, and Validation benchmarks CytePool against the reference packages.

Documentation

The documentation is at https://docs.nygen.io/CytePool/.

Development

uv sync --extra test --group dev --group docs
uv run pytest
uv run ruff check cytepool tests validation docs
uv run ruff format --check cytepool tests validation docs
uv run mypy cytepool
make -C docs api

License

BSD 3-Clause. See LICENSE and THIRD_PARTY_NOTICES.md.