Strategy reference

PeakATail is built around pluggable strategy registries. Every algorithm choice — how cells are clustered, how APA usage is quantified, how differential tests are run, how clusters are matched across datasets, and which figures are produced — is a named strategy that can be swapped without touching the rest of the pipeline. This page lists every registered strategy and links to its full reference.

Clustering

Clustering groups cells by their 3' UTR usage so that downstream APA tests have biologically meaningful contrasts. See clustering strategies.

Strategy	CLI flag	When to use
leiden_tfidf	--cluster-method leiden_tfidf	Sparse PAS counts (10x Chromium, Drop-seq). Recommended default.
leiden_libsize	--cluster-method leiden_libsize	Dense gene-count-like data; when TF-IDF produces too many small clusters.
external	--cluster-method external	Import Seurat / Scanpy clusters produced from gene expression.

Quantification (PDUI)

Quantification converts the PAS count matrix into a per-cell APA summary score. See quantification strategies.

Strategy	CLI flag	Output file	When to use
classic	--pdui-method classic	pdui_classic.tsv	Binary proximal/distal question on a per-gene or per-isoform basis.
proportion	--pdui-method proportion	proportion.tsv	Full per-PAS usage vector per cell; feeds proportion_heatmap viz.
shannon	--pdui-method shannon	entropy_shannon.tsv	Heterogeneity of PAS usage; genes with many PAS, or pseudotime analyses.

Differential APA

Differential testing identifies PAS that shift between two or more clusters. See diff strategies.

Strategy	CLI flag	Multi-condition	When to use
fisher	--diff-method fisher	No	Fast exploratory screen; within-gene framing (DEXSeq-style).
nb_pairwise	--diff-method nb_pairwise	No	Cell-level NB GLM; corrects for library size; no pseudo-replication.
nb_multi	--diff-method nb_multi	Yes	Omnibus test across all clusters at once.

Cross-dataset matching

Matching assigns canonical cluster IDs when the same experiment is processed in multiple batches or from multiple samples. See match strategies.

Strategy	CLI flag	When to use
marker_overlap	--match-method marker_overlap	Default. Datasets share PAS features; reasonably similar protocols.
mnn	--match-method mnn	Low marker overlap; different capture technologies or library sizes.
jaccard	--match-method jaccard	Same physical cells in both runs (regression testing or split-BAM experiments).

Visualization

Every plot type is a registered strategy with optional engine variants (matplotlib, plotly, scanpy). See viz strategies.

Plot type	Registered names	What it shows
umap	umap_matplotlib, umap_plotly, umap_scanpy	2-D UMAP colored by cluster.
cluster_sizes	cluster_sizes_matplotlib, cluster_sizes_plotly	Bar chart of cell counts per cluster.
peak_qc	peak_qc_matplotlib, peak_qc_plotly	4-panel peak and cell QC distributions.
volcano	volcano_matplotlib, volcano_plotly	Differential APA volcano: log2fc vs -log10(q).
pdui_distribution	pdui_distribution_matplotlib, pdui_distribution_plotly, pdui_distribution_scanpy	Per-cluster PDUI violin plots.
proportion_heatmap	proportion_heatmap_matplotlib, proportion_heatmap_plotly	PAS x cluster mean proportion heatmap.
entropy_distribution	entropy_distribution_matplotlib, entropy_distribution_plotly	Per-cluster Shannon entropy violin plots.
diff_agreement	diff_agreement_matplotlib, diff_agreement_plotly	Jaccard overlap between significant PAS sets from different test strategies.
length_shifts	length_shifts_matplotlib, length_shifts_plotly	Gene x cluster-pair PDUI delta heatmap.
pas_overlap	pas_overlap_matplotlib, pas_overlap_plotly	UpSet / Venn of PAS detected across datasets.
atlas_snap_diag	atlas_snap_diag_matplotlib, atlas_snap_diag_plotly	Atlas-snap result: snapped vs unsnapped counts + distance histogram.
cluster_match_sankey	cluster_match_sankey_matplotlib, cluster_match_sankey_plotly	Alluvial chart of original → canonical cluster correspondence.
match_confidence	match_confidence_matplotlib, match_confidence_plotly	Heatmap of match confidence scores per (dataset, cluster) pair.
tile_timing	tile_timing_matplotlib, tile_timing_plotly	Bar chart of per-tile peak-calling wall time, outliers highlighted.
resource_timeline	resource_timeline_matplotlib, resource_timeline_plotly	Dual-axis RAM (GB) and CPU% over elapsed seconds.
gene_track	gene_track_matplotlib, gene_track_plotly	Per-gene: isoform structure + per-cluster PAS coverage bars.