API Map¶

RustScenic exposes the workflow as small Python calls. Each stage can be used alone, or combined through rustscenic.pipeline.run.

RNA Regulatory Network¶

rustscenic.grn.infer(adata, tf_names, n_estimators=500, seed=777)

Returns a pandas.DataFrame with transcription factor, target and importance columns.

Use this when you need a GRNBoost2-style TF-target edge table without depending on arboreto and dask at runtime. For very high cell counts, target_block_size=None uses the adaptive target-blocking default; pass a positive integer only when benchmarking a specific cache/RSS tradeoff.

AUCell¶

rustscenic.aucell.score(adata, regulons, top_frac=0.05)

Returns a cells by regulons activity matrix.

Use this when you already have regulons and need per-cell TF programme activity.

cisTarget¶

rustscenic.cistarget.enrich(rankings, regulons, nes_threshold=3.0)

Returns motif enrichment rows with AUC and NES values.

Use this for motif support filtering of candidate regulons.

Topics¶

rustscenic.topics.fit(atac_adata, n_topics=30)
rustscenic.topics.fit_gibbs(atac_adata, n_topics=30, n_threads=8)

Use Online VB for smaller or faster exploratory runs. Use collapsed Gibbs for higher topic diversity at larger K.

ATAC Preprocessing¶

rustscenic.preproc.fragments_to_matrix("fragments.tsv.gz", "peaks.bed")

Returns an AnnData peak matrix suitable for topic modelling.

Pipeline¶

rustscenic.pipeline.run(rna=adata, tfs=tfs, output_dir="out")

Use the orchestrator when you want the full staged workflow and a manifest.

When region_motif_rankings points to a parquet or feather file, the orchestrator reads only the motif ID column plus the peak columns used by the current run. This keeps large region-ranking databases from being loaded in full. Peak IDs in the BED or ATAC matrix must match the ranking database region IDs.