scTenifoldXct

A semi-supervised method for predicting cell-cell interactions and mapping cellular communication graphs via manifold learning. [Paper]

🆕 New: a local web UI. No code required:

pip install --no-cache-dir "scTenifoldXct[web]"
sctenifoldxct-ui
# opens http://127.0.0.1:8765

Load a dataset (the bundled example needs no upload, from a git checkout — otherwise upload your own .h5ad), pick sender/receiver cell types, and hit Run analysis. Ranked ligand-receptor pairs show up in-browser with a full CSV download. Everything — data, computation, results — stays on your machine. Run sctenifoldxct-ui --help for host/port/GRN-cache options.

Prefer a hosted, always-on option instead? Try the Streamlit demo.

Install

Requires Python 3.10+.

pip install scTenifoldXct

From source (development or latest unreleased changes):

git clone https://github.com/cailab-tamu/scTenifoldXct.git
cd scTenifoldXct
pip install .

Example Data

Two real, ready-to-use datasets are bundled under data/ — no data of your own required:

Both are log-normalised and ready to feed straight into the examples below. See data/README.md for additional/larger datasets.

Usages

Quick Start

import logging
import scanpy as sc
import scTenifoldXct as st

# scTenifoldXct logs progress via the logging module; configure a handler to
# see messages when verbose=True (e.g. in a script or notebook):
logging.basicConfig(level=logging.INFO, format="%(message)s")

adata = sc.read_h5ad('data/adata_short_example.h5ad') # load data
xct = st.scTenifoldXct(data = adata, # an AnnData
                    source_celltype = 'Inflam. FIB', # sender cell type
                    target_celltype = 'Inflam. DC', # receiver cell type
                    obs_label = 'ident', # colname in adata.obs indicating cell types
                    rebuild_GRN = True, # whether to build GRNs
                    GRN_file_dir = 'Net_example_dev',  # folder path to GRNs
                    verbose = True, # whether to log the processing
                    n_cpus = -1) # CPU multiprocessing, -1 to use all
emb = xct.get_embeds(train = True) # Manifold alignment to project data to low-dimensional embeddings
xct_pairs = xct.null_test() # non-parametric test to get significant interactions
print(xct_pairs)

Command line

# single-sample interaction analysis
sctenifoldxct data/adata_short_example.h5ad --rebuild \
    -s "Inflam. FIB" -r "Inflam. DC" --n_cpus 8 -v

# two-sample differential interaction analysis
sctenifoldxct-merge data/adata_merge_example.h5ad NormalvsTumor N T \
    --rebuild -s "B cells" -r "Fibroblasts" --n_cpus 8 -v

Run sctenifoldxct --help or sctenifoldxct-merge --help for all options.

Tutorial

Two tutorial notebooks cover usage and results visualization:

• Single-sample interaction analysis: https://github.com/cailab-tamu/scTenifoldXct/blob/main/tutorials/tutorial-short_example.ipynb
• Two-sample differential interaction analysis: https://github.com/cailab-tamu/scTenifoldXct/blob/main/tutorials/tutorial-merge_short_example.ipynb

Run scTenifoldXct from command-line by Docker

A Docker image with all required packages and databases included, for users not familiar with Python:

docker build -t sctenifold .
docker run -it --name xct --shm-size=8gb sctenifold

This drops you into a Bash terminal in the container. Example single-sample run:

sctenifoldxct data/adata_short_example.h5ad \
--rebuild \
-s "Inflam. FIB" \
-r "Inflam. DC" \
--n_cpus 8 \
-v

Two-sample analysis:

sctenifoldxct-merge data/adata_merge_example.h5ad \
NormalvsTumor N T \
--rebuild \
-s "B cells" \
-r "Fibroblasts" \
--n_cpus 8 \
-v

Copy your own data into the container for your analyses. When done, detach with Ctrl+p Ctrl+q and copy results back to the host:

docker cp xct:/app/xct_results/ .