Node reference

Analysis nodes.

Dimensionality reduction, batch correction, the kNN graph, embeddings, clustering and differential expression. Clustering offers Leiden, Louvain and — since v0.24.0 — K-Means with an exact number of clusters; all three produce labels that behave identically downstream. DEG Analysis ranks every gene per group (Wilcoxon by default, t-test available).

08 nodesCategory · analysisin 17 templates

Recommended order

PCA → (Harmony) → Neighbors → UMAP → Clustering → DEG Analysis. After Harmony, set use_rep='X_pca_harmony' in Neighbors. Scale saves adata.raw automatically, which is what DEG Analysis reads for accurate pct values.

K-Means (method='kmeans', n_clusters) partitions the same PCA / Harmony embedding; it does not use resolution, and the generated Methods text omits it accordingly.

Common pitfalls

Harmony writes X_pca_harmony, not X_pca. Unless Neighbors is told to use it, UMAP and clustering run on the uncorrected PCA.

UMAP without Neighbors. UMAP needs the kNN graph; the required order is PCA → Neighbors → UMAP. The node’s panel warns when the graph is missing upstream.

DEG pct values all 1.0. Run Scale before DEG Analysis so adata.raw (pre-scale, log-normalised) exists for pct_nz_group / pct_nz_reference.

DEG needs at least two populated groups. A single-group column stops with a clear message instead of a division-by-zero error.

UMAP distance is not biological distance. Only local neighbourhood structure is preserved; use PAGA for trajectory topology.

PCA

pca

Principal Component Analysis — required before Neighbors/UMAP/Harmony.

Node LibraryscRNA-seqSpatial

InAnnData (scaled)

OutAnnData (obsm["X_pca"], varm["PCs"])

Parameters (7)

Name	Type	Default	Description
`n_comps`	number	`50`	Number of PCs.
`use_highly_variable`	checkbox	`true`	Use HVGs only (recommended).
`layer`	layer	—	Data layer to use (empty = X matrix).
`svd_solver`	select	`arpack`	"arpack" exact, "randomized" fast.
`random_state`	number	`0`	Random seed.
`chunked`	checkbox	`false`	Process in chunks for very large datasets.
`chunk_size`	number	`20000`	Number of cells per chunk.

Neighbors

neighbors

k-nearest neighbor graph on PCA space. Required for clustering and UMAP.

Node LibraryscRNA-seqSpatialATAC-seq

InAnnData (with PCA)

OutAnnData (obsp connectivities/distances)

Parameters (6)

Name	Type	Default	Description
`n_neighbors`	number	`15`	10-15 fine, 30-50 smooth.
`n_pcs`	number	`15`	Default 15. 0 = use all PCs.
`use_rep`	select	`auto`	"X_pca_harmony" after Harmony.
`metric`	select	`euclidean`	euclidean/cosine/manhattan/correlation.
`method`	select	`umap`	Graph construction method: 'umap' (default), 'gauss', 'rapids' (requires rapids-singlecell + GPU).
`random_state`	number	`0`	Random seed.

UMAP

umap

UMAP 2D embedding for visualization. Local neighborhood structure preserved; global distances are NOT meaningful.

Node LibraryscRNA-seqSpatialATAC-seq

InAnnData (with neighbors)

OutAnnData (obsm["X_umap"])

Parameters (8)

Name	Type	Default	Description
`min_dist`	number	`0.1`	Lower = tighter clusters.
`spread`	number	`1.0`	Spread of embedded points.
`n_components`	number	`2`	2 or 3.
`init_pos`	select	`spectral`	spectral/paga (trajectory)/random.
`alpha`	number	`1.0`	Initial learning rate (slider 0.1 .. 5). Default 1.0 is fine for most cases; very rarely tuned.
`gamma`	number	`1.0`	Weighting of negative samples (slider 0.1 .. 5). Default 1.0 is fine.
`random_state`	number	`0`	Random seed.
`maxiter`	number	`0`	Maximum number of iterations. 0 = Scanpy automatic default.

t-SNE

tsne

t-SNE embedding (alternative to UMAP). Good for distinct clusters; slower than UMAP.

Node LibraryscRNA-seq

InAnnData (with PCA)

OutAnnData (obsm["X_tsne"])

Parameters (6)

Name	Type	Default	Description
`perplexity`	number	`30`	5-50; matches expected cluster size.
`learning_rate`	number	`1000`	Usually N/12.
`n_pcs`	number	`0`	Number of PCA components to use. 0 = automatic.
`use_rep`	select	`auto`	"X_pca_harmony" after Harmony.
`early_exaggeration`	number	`12.0`	Controls cluster separation in early iterations.
`random_state`	number	`0`	Random seed.

Clustering

clustering

Leiden or Louvain community detection on the kNN graph, or K-Means with a fixed number of clusters on the embedding.

Node LibraryscRNA-seqSpatialATAC-seq

InAnnData (with neighbors)

OutAnnData (obs["leiden"] or obs["louvain"])

Parameters (9)

Name	Type	Default	Description
`method`	select	`leiden`	"leiden" or "louvain".
`resolution`	number	`1.0`	0.3 coarse, 1.0 standard, 2.0 fine.
`n_clusters`	number	`8`	Exact number of clusters. Used by method='kmeans' only — ignored by leiden/louvain.
`key_added`	text	`leiden`	obs column name (default = method).
`flavor`	select	`leidenalg`	Leiden implementation: 'leidenalg' (default) or 'igraph'.
`n_iterations`	number	`-1`	Iterations for Leiden (-1 = until convergence).
`kmeans_n_init`	number	`10`	Number of random K-Means restarts, keeping the best inertia. method='kmeans' only.
`random_state`	number	`0`	Random seed.
`restrict_to`	text	—	obs column to sub-cluster within.

Edit Clusters

editClusters

Rename or merge cluster labels for cell type annotation. Mapping format: "old1, old2: new_label" (one rule per line). Use after identifying cell types from marker genes.

Node LibraryscRNA-seqSpatial

InAnnData (with clusters)

OutAnnData (new obs column)

Parameters (4)

Name	Type	Default	Description
`source_column`	obs column	`leiden`	Existing cluster column (e.g. "leiden").
`new_column`	text	`cell_type`	New column name.
`mapping`	label map	—	Merge/rename rules.
`categories_order`	text	—	Custom order (comma-separated).

DEG Analysis

findMarkerssince v0.9.0-beta.25 (renamed from Find Markers)

Differential expression — ranks all genes per cluster (Wilcoxon by default). Replaces "Find Markers" naming. n_genes=None ranks ALL genes. pts=True computes pct_nz_group / pct_nz_reference (requires adata.raw, set automatically by Scale).

Node LibraryscRNA-seq

InAnnData (with clusters)

OutAnnData (uns["rank_genes_groups"])

Parameters (12)

Name	Type	Default	Description
`groupby`	obs column	`leiden`	Group label column.
`method`	select	`wilcoxon`	wilcoxon/t-test/t-test_overestim_var/logreg.
`n_genes`	number	—	null = all genes (recommended).
`reference`	text	`rest`	"rest" or specific group for 1-vs-1.
`groups`	text	—	Restrict the DEG test to specific groups of groupby (comma-separated, e.g. '0,2' or 'Normal'). Empty (default) tests every group against the reference.
`use_raw`	checkbox	`true`	Use adata.raw (set by Scale).
`layer`	layer	—	Expression layer to use instead of X/raw.
`corr_method`	select	`benjamini-hochberg`	"benjamini-hochberg" or "bonferroni".
`tie_correct`	checkbox	`false`	Correct for ties in Wilcoxon test.
`rankby_abs`	checkbox	`false`	Rank by absolute value of score.
`pts`	checkbox	`true`	Compute pct_nz_group / pct_nz_reference.
`key_added`	text	`rank_genes_groups`	Key in adata.uns to store results.

Harmony

harmony

Harmony batch correction — outputs to obsm["X_pca_harmony"]. Set use_rep="X_pca_harmony" in Neighbors after.

Node LibraryscRNA-seqSpatial

InAnnData (with PCA)

OutAnnData (obsm["X_pca_harmony"])

Parameters (10)

Name	Type	Default	Description
`batch_key`	obs column	`batch`	obs column with batch labels.
`theta`	number	`2.0`	Diversity penalty (correction strength).
`basis`	select	`X_pca`	Input embedding key in adata.obsm to correct. Only 'X_pca' is currently supported (PCA must be run upstream).
`adjusted_basis`	text	`X_pca_harmony`	Output embedding name after correction.
`sigma`	number	`0.1`	Soft assignment width.
`nclust`	number	`0`	Number of clusters for Harmony. 0 = auto.
`max_iter_harmony`	number	`10`	Max Harmony iterations.
`max_iter_kmeans`	number	`20`	Maximum K-means iterations per Harmony round.
`block_size`	number	`0.05`	Block size for ridge regression.
`random_state`	number	`0`	Random seed.

Preprocessing

Trajectory