RustScenic is built to be the simplest, fastest way to run regulatory-network analysis in single-cell data. It brings core pySCENIC and SCENIC+ analysis stages into a Python package powered by Rust, developed with the Huang Lab at Mount Sinai.

For researchers dealing with installation, memory or runtime problems in existing workflows, a useful first step is to inspect the outputs. RustScenic provides a saved-network browser example, an offline example, and a notebook for running a fresh analysis on public data.

Inspect a network without installing anything

Open the endothelial-cell example and use the Regulator selector. The default is ETS1; changing it to ERG updates the predicted-target table and the gene-set summary.

The example contains a cohort of 5,000 sampled cells drawn from 234 source datasets, with 184,000 retained network edges and 85 motif-supported regulons. The table shows target genes, model importance and expression correlation. It displays the leading targets within the selected gene set.

These are exploratory predictions. Model importance is not a probability, correlation is not causation, and motif support for a gene set does not experimentally verify every interaction. The cohort pools tissue and disease annotations. Source-study citations and provenance are linked beneath the viewer.

The viewer reads saved results. Selecting another regulator does not fit a model or upload an expression matrix.

Explore the same results offline

Download the offline starter ZIP, extract it, and open explore.html in a browser.

With Python 3.10 or later, the extracted example also supports:

python3 example.py --verify
python3 example.py --tf ERG

Run these commands from the extracted folder. They verify or inspect the bundled results; they do not download a new dataset or run inference. The file guide describes the network tables, motif evidence, activity summaries and checksums.

Run a fresh RNA analysis in Python

The PBMC3k notebook installs the published package and downloads public 10x data. It walks through input preparation, transcription-factor-to-target inference, candidate gene-set activity scoring and result export.

For local use, the quickstart uses regular CPython 3.10–3.14 and the examples dependencies:

python -m pip install "rustscenic[examples]"

The walkthrough writes network edges to grn.parquet, activity scores to aucell.parquet, the prepared RNA matrix to pbmc3k.h5ad, and the matching transcription-factor names to tfs.txt.

This small teaching workflow uses limited gene and transcription-factor subsets. Its candidate sets are not motif-filtered or separated by correlation sign, and the notebook does not establish biological accuracy. The PBMC3k analysis and the saved endothelial-cell example use different data and serve different purposes.

Prepare for your own study

For the documented own-data RNA path, use a cells-by-genes AnnData matrix with log-normalised expression in adata.X and gene symbols in adata.var_names. Match transcription-factor identifiers and species to the matrix. Motif refinement and multiome stages require their own compatible inputs; see the API and input guides.

The repository publishes performance measurements with dataset sizes, hardware and timing boundaries. Agreement with reference tools varies by stage; full scientific output parity is not claimed. Review those conditions and the validation notes before replacing a study workflow.

Start at the GitHub repository, try the saved example, and use the notebook to evaluate the Python workflow. Feedback on installation, AnnData handling and output interpretation will help guide the project.

Maintained by Ekin Kahraman. Shared on behalf of the RustScenic project. AI assistance was used for drafting and source checks. The browser example and regulator selection were inspected while preparing this article; no new performance or biological-validation experiment was performed.