Node reference

Preprocessing nodes.

Standard scRNA-seq preprocessing. Quality Control is one node that computes the metrics, opens the interactive QC Workspace (MAD-based recommendations, Raw Data Mode, thresholds that update the plots live without re-running) and applies the filters. Then Scrublet doublet detection, normalization, log transform, highly variable gene selection and scaling.

06 nodesCategory · preprocessin 18 templates

Recommended order

Quality Control → (Doublet Detection) → Normalize → Log1p → Highly Variable Genes → Scale → PCA. Open the QC Workspace from the node card before committing to thresholds — see the QC Workspace page.

Scale keeps hvg_only on by default: X keeps only the highly variable genes, the full gene set is preserved in adata.raw for DEG and gene plots, and the dense-matrix memory cost drops by roughly an order of magnitude on large data.

Common pitfalls

Wrong pipeline order. Never run Scale before Normalize — scaling raw counts produces meaningless PCA results.

Mouse mito prefix. Mouse uses lowercase mt-, human uses MT-. The wrong prefix gives 0% mitochondrial reads for every cell.

After Scale, adata.X holds z-scores, not expression. Plot nodes default to use_raw, which reads the log-normalised values Scale preserved; if you switch it off, pick an expression layer explicitly.

Keep save_counts on in Normalize. It stores raw counts in layers["counts"], which count-based steps downstream rely on.

Check the QC filter list on the node card. Since v0.24.0 the card lists exactly which filters will run; a Quality Control node edited in older versions through the workspace may have had Max Genes / Max Counts switched on unintentionally.

Quality Control

qualityControlsince v0.9.0-beta.29

Filter cells/genes by QC metrics. Unified node (replaces qcMetrics + qcVisualization). Use the QC Workspace toolbar button for interactive threshold exploration. Each filter has its own enable toggle. Built-in plot replaces the old separate qcVisualization node.

Node LibraryscRNA-seqSpatial

InAnnData

OutAnnData (filtered, QC obs columns)

Parameters (33)

Name	Type	Default	Description
`show_plot`	checkbox	`true`	Generate built-in QC plot.
`plot_type`	select	`violin`	violin/scatter/histogram/highest_expr_genes/scatter_matrix.
`color_by`	obs column	—	Obs column to color cells by (empty = auto QC_label coloring — PASS/FAIL).
`group_colors`	color map	—	Canonical categorical color overrides for the active color_by owner. Empty color_by uses obs['QC_label']; null preserves the existing plot palette.
`layer`	text	—	Data layer to use for QC calculation (empty = X matrix).
`mito_prefix`	text	`MT-`	"MT-" human, "mt-" mouse.
`ribo_prefix`	text	`RPS,RPL`	Ribosomal gene prefixes.
`hb_prefix`	text	`HB`	Hemoglobin gene prefix.
`filter_min_genes`	checkbox	`true`	Enable minimum genes per cell filter.
`min_genes`	number	`200`	Min genes per cell.
`filter_max_genes`	checkbox	`false`	Enable maximum genes per cell filter.
`max_genes`	number	`5000`	Max genes per cell (doublet filter).
`filter_min_counts`	checkbox	`false`	Enable minimum counts per cell filter.
`min_counts`	number	`500`	Minimum total UMI counts per cell.
`filter_max_counts`	checkbox	`false`	Enable maximum counts per cell filter.
`max_counts`	number	`50000`	Maximum total UMI counts per cell.
`filter_max_pct_mito`	checkbox	`true`	Enable mitochondrial percentage filter.
`max_pct_mito`	number	`20`	Max mito percentage.
`filter_max_pct_ribo`	checkbox	`false`	Enable ribosomal percentage filter.
`max_pct_ribo`	number	`50`	Max ribosomal percentage.
`filter_max_pct_hb`	checkbox	`false`	Enable hemoglobin percentage filter.
`max_pct_hb`	number	`5`	Maximum hemoglobin gene percentage.
`filter_min_cells`	checkbox	`true`	Enable minimum cells per gene filter.
`min_cells`	number	`3`	Min cells expressing a gene.
`raw_data_mode`	checkbox	`false`	For raw/unfiltered droplet data (Cellranger raw).
`n_mads`	number	`3`	MAD multiplier for recommended thresholds (median ± n_mads × MAD). Only applied when raw_data_mode is ON; otherwise standard MAD×5 is used. Lower values (e.g. 2) are more aggressive at removing outliers; higher values (e.g. 5) are more permissive. Tunable inline in the QC Workspace modal header; changes update the `rec:` buttons instantly (recommendations are recomputed client-side from cell_data, no backend re-run). Click 'Auto-fill MAD recommendations' to apply the rec values to the actual threshold sliders. Valid range 1–10.
`clip_percentile`	number	`1`	Lower percentile used to clip QC Workspace plot axes when raw_data_mode is ON. Value 1 means axes are clipped to [1, 99] percentile. Set to 0 to disable clipping even in Raw Mode. Ignored entirely when raw_data_mode is OFF. Tunable inline in the modal header; Plotly axes re-clip instantly via useMemo (no backend re-run). Valid range 0–10. Note: this only affects the modal view — it does not modify filter values or the underlying data.
`percent_top`	text	—	Calculate % counts from top N genes as a JSON array of integers, e.g. [50, 100, 200]. Omit or null to skip.
`log1p_metrics`	checkbox	`false`	Calculate log1p-transformed QC metrics.
`metrics`	array	`["n_genes_by_counts", "total_counts", "pct_counts_mt"]`	JSON array of QC metric names to include in the built-in plot. Omit this parameter to use the default metrics.
`log_scale`	checkbox	`false`	Use log scale for axes in histogram/scatter plots.
`font_size`	text	—	Font size for plot labels. null = matplotlib default.
`publication_quality`	checkbox	`false`	Generate high-resolution (300 DPI) publication-quality figure.

Normalize

normalize

Normalize total counts per cell (scanpy normalize_total).

Node LibraryscRNA-seqSpatial

InAnnData

OutAnnData (normalized X, layers["counts"])

Parameters (5)

Name	Type	Default	Description
`target_sum`	number	`10000`	Target counts per cell (1e4 = CP10K).
`save_counts`	checkbox	`true`	Save raw counts to layers["counts"].
`layer`	layer	—	Layer to normalize (empty = X matrix).
`exclude_highly_expressed`	checkbox	`false`	Exclude top genes from normalization factor.
`max_fraction`	number	`0.05`	Threshold for "highly expressed".

Log1p

log1p

log(x+1) transformation. Run after Normalize, before HVG/PCA.

Node LibraryscRNA-seqSpatial

InAnnData (normalized)

OutAnnData (log-transformed X)

Parameters (5)

Name	Type	Default	Description
`save_normalized`	checkbox	`true`	Save normalized+log1p to layers["normalized"].
`layer`	log1p_layer_select	—	Layer to transform (empty = X matrix).
`base`	log1p_base_select	—	null = natural log, 2/10 for log2/log10.
`chunked`	log1p_chunked_checkbox	`false`	Process in chunks for very large datasets.
`chunk_size`	log1p_chunk_size	`10000`	Number of elements per chunk.

Highly Variable Genes

highlyVariableGenes

Identify highly variable genes (HVGs) that drive biological variation.

Node LibraryscRNA-seqSpatial

InAnnData (log-transformed)

OutAnnData (var["highly_variable"])

Parameters (10)

Name	Type	Default	Description
`n_top_genes`	number	`2000`	1000-3000 typical.
`flavor`	select	`seurat`	"seurat" log, "seurat_v3" raw counts, "cell_ranger".
`layer`	layer	—	Data layer to use.
`batch_key`	obs column	—	obs column for batch-aware selection.
`subset`	checkbox	`false`	Directly subset adata to HVGs.
`min_mean`	number	`0.0125`	Minimum mean expression cutoff (seurat/cell_ranger flavors).
`max_mean`	number	`3`	Maximum mean expression cutoff.
`min_disp`	number	`0.5`	Minimum dispersion cutoff.
`span`	number	`0.3`	LOESS smoothing span (seurat_v3 flavor only).
`n_bins`	number	`20`	Number of expression bins.

Scale

scale

Scale to unit variance and zero-center. Required before PCA. Auto-saves adata.raw.

Node LibraryscRNA-seqSpatial

InAnnData

OutAnnData (X scaled, adata.raw set)

Parameters (4)

Name	Type	Default	Description
`max_value`	number	`10`	Clip values above this.
`zero_center`	checkbox	`true`	Subtract mean (centering).
`hvg_only`	checkbox	`true`	Scale only the highly-variable genes (requires highlyVariableGenes upstream). This is the standard scanpy workflow and the default. Cuts the dense-matrix RAM blow-up ~10x on large data; the full gene set is kept in adata.raw for DEG/plots, and X keeps only the HVG.
`save_log1p`	checkbox	`true`	Save pre-scale log1p to layers["log1p"].

Doublet Detection

doubletDetection

Detect doublets using Scrublet. Run early in the pipeline.

Node LibraryscRNA-seq

InAnnData

OutAnnData (obs["predicted_doublet"], obs["doublet_score"])

Parameters (6)

Name	Type	Default	Description
`mode`	select	`filter`	"filter" removes; "label" only marks.
`expected_doublet_rate`	number	`0.06`	~0.008 × cells/1000.
`threshold`	number	—	null = auto-detect.
`n_prin_comps`	number	`30`	PCs for simulated doublets.
`min_gene_variability_pctl`	number	`85`	Percentile for gene variability filtering.
`batch_size`	number	`100000`	Maximum number of observed cells processed in each independent Scrublet batch. Datasets larger than this are split into consecutive batches; default 100,000.

Input / Output

Analysis