scTenifoldpy

scTenifoldpy is a Python implementation of the scTenifold suite: scTenifoldNet (cross-condition GRN comparison), scTenifoldKnk (virtual knockout), and scTenifoldXct (cell-cell interaction prediction, provided via the separately maintained scTenifoldXct package).

Local Web UI

Run scTenifoldNet, scTenifoldKnk, or plain GRN construction from a browser instead of code:

pip install "scTenifoldpy[ui]"
sctenifold-ui

sctenifold-ui defaults to port 8001 (not the more commonly used 8000, to lower the odds of colliding with another local server) and opens http://127.0.0.1:8001 in your browser. Use --port to pick a different one.

This opens a local page (http://127.0.0.1:8001) where you can try a bundled synthetic dataset or the real 10x PBMC3k dataset (the one behind the Seurat/Scanpy tutorials), or upload your own data as a genes-by-cells CSV or an AnnData .h5ad file. Pick a workflow, run it, and download the ranked-gene or ranked-edge results as CSV. Everything runs locally; no data leaves your machine. See the UI guide for details.

Installation

uv venv
uv add scTenifoldpy

or:

pip install scTenifoldpy

Optional extras:

# use uv
uv venv
uv add "scTenifoldpy[scanpy]"
uv add "scTenifoldpy[parallel-ray]"
uv add "scTenifoldpy[ui]"

# or use pip
pip install "scTenifoldpy[scanpy]"
pip install "scTenifoldpy[parallel-ray]"
pip install "scTenifoldpy[ui]"

The scTenifoldXct cell-cell interaction workflow is shipped as a separately maintained dependency behind the xct extra (pulls PyTorch and scanpy; requires Python >= 3.10):

pip install "scTenifoldpy[xct]"

Docker

Build the default runtime image:

docker build -t sctenifoldpy .

Run the CLI from the container, mounting the current directory as the working directory:

docker run --rm -v "$PWD:/workspace" sctenifoldpy scTenifold --help

PowerShell:

docker run --rm -v "${PWD}:/workspace" sctenifoldpy scTenifold --help

Optional extras can be included at build time:

docker build --build-arg EXTRAS=scanpy -t sctenifoldpy:scanpy .
docker build --build-arg EXTRAS=parallel-ray -t sctenifoldpy:ray .

Class API

from scTenifold.data import get_test_df
from scTenifold import scTenifoldNet

df_1 = get_test_df(n_cells=1000)
df_2 = get_test_df(n_cells=1000)

sc = scTenifoldNet(
    df_1,
    df_2,
    "X",
    "Y",
    qc_kws={"min_lib_size": 10},
    nc_kws={"backend": "serial", "n_jobs": 1},
)
result = sc.build()

High-Level API

from scTenifold import compare_networks, virtual_knockout

result = compare_networks(
    df_1,
    df_2,
    qc_kws={"min_lib_size": 10, "plot": False},
    network_kws={"n_nets": 3, "n_samp_cells": 100},
    backend="joblib-threading",
    n_jobs=4,
)

knockout = virtual_knockout(
    df_1,
    ko_genes=["NG-1"],
    qc_kws={"min_lib_size": 10, "min_percent": 0.001},
)

Cell-Cell Interaction (scTenifoldXct)

scTenifoldXct is maintained separately (cailab-tamu/scTenifoldXct) and re-exported here for convenience — from scTenifold import scTenifoldXct returns the exact same class and results:

import scanpy as sc
from scTenifold import scTenifoldXct

adata = sc.read_h5ad("log_normalised.h5ad")
xct = scTenifoldXct(
    data=adata,
    source_celltype="cell_A",
    target_celltype="cell_B",
    obs_label="ident",
    rebuild_GRN=True,
    GRN_file_dir="./xct_results",
)
xct.get_embeds(train=True)
enriched = xct.chi2_test()

See the scTenifoldXct guide for the CLI and differential (two-sample) analysis. Requires scTenifoldpy[xct].

Parallel Backends

Network construction defaults to deterministic serial execution:

from scTenifold import make_networks

networks = make_networks(df_1, backend="serial", n_jobs=1)
networks = make_networks(df_1, backend="joblib-loky", n_jobs=4)

Supported backends are serial, joblib-loky, joblib-threading, and ray. Ray is optional and requires scTenifoldpy[parallel-ray].

CLI

scTenifold config -t 1 -p ./net_config.yml
scTenifold net -c ./net_config.yml -o ./output_folder
scTenifold knk -c ./knk_config.yml -o ./output_folder
scTenifold xct data.h5ad -s cell_A -r cell_B -l ident   # needs [xct]

Citation

If you use scTenifoldpy in scientific work, please cite this software using the metadata in CITATION.cff, and cite the underlying method paper that matches your analysis:

scTenifoldNet

Osorio, D., Zhong, Y., Li, G., Huang, J. Z., & Cai, J. J. (2020). scTenifoldNet: A machine learning workflow for constructing and comparing transcriptome-wide gene regulatory networks from single-cell data. Patterns, 1(9), Article 100139. https://doi.org/10.1016/j.patter.2020.100139

scTenifoldKnk

Osorio, D., Zhong, Y., Li, G., Xu, Q., Yang, Y., Tian, Y., Chapkin, R. S., Huang, J. Z., & Cai, J. J. (2022). scTenifoldKnk: An efficient virtual knockout tool for gene function predictions via single-cell gene regulatory network perturbation. Patterns, 3(3), Article 100434. https://doi.org/10.1016/j.patter.2022.100434

scTenifoldXct

Yang, Y., Osorio, D., Long, W., Bai, M., Wang, F., Yan, X., Yang, S., Chen, A., Zhang, P., Cai, J. J. (2023). scTenifoldXct: A semi-supervised method for predicting cell-cell interactions and mapping cellular communication graphs. Cell Systems, 14(4), 302-311. https://doi.org/10.1016/j.cels.2023.01.004