WNN integration function

Combine paired RNA and ATAC representations with weighted nearest neighbors, retaining both modality reductions and the WNN graph.

Usage

WNN_integrate(
  srt_merge = NULL,
  batch = NULL,
  append = TRUE,
  srt_list = NULL,
  assay = NULL,
  do_normalization = NULL,
  normalization_method = "LogNormalize",
  do_HVF_finding = TRUE,
  HVF_source = "separate",
  HVF_method = "vst",
  nHVF = 2000,
  HVF_min_intersection = 1,
  HVF = NULL,
  do_scaling = TRUE,
  vars_to_regress = NULL,
  regression_model = "linear",
  scale_within_batch = FALSE,
  linear_reduction = "pca",
  linear_reduction_dims = 50,
  linear_reduction_dims_use = NULL,
  linear_reduction_params = list(),
  force_linear_reduction = FALSE,
  nonlinear_reduction = "umap",
  nonlinear_reduction_dims = c(2, 3),
  nonlinear_reduction_params = list(),
  force_nonlinear_reduction = TRUE,
  neighbor_metric = "euclidean",
  neighbor_k = 20L,
  cluster_algorithm = "louvain",
  cluster_resolution = 0.6,
  verbose = TRUE,
  seed = 11
)

Arguments

srt_merge: A merged `Seurat` object that includes the batch information.
batch: Accepted for integration-interface compatibility. WNN does not perform batch correction; supplying this argument emits a warning.
append: Append integrated results to srt_merge.
srt_list: A list of Seurat objects to be checked and preprocessed.
assay: One RNA assay and one chromatin assay, or one assay whose partner can be inferred. Automatic selection prefers an active RNA/SCT assay, then RNA, then SCT; custom RNA assay names must be explicit. Known SCOP gene-activity and chromVAR assays cannot substitute for RNA.
do_normalization: Whether data normalization should be performed.
normalization_method: The normalization method to be used. Possible values are "LogNormalize", "SCT", "TFIDF", and "scran".
do_HVF_finding: Whether to select variable features during preprocessing.
HVF_source: The source of highly variable features. Possible values are "global" and "separate".
HVF_method: RNA feature-selection method passed to RunStandardWorkflow().
nHVF: Maximum number of selected features, subject to modality defaults and the number of available features.
HVF_min_intersection: The feature needs to be present in batches for a minimum number of times in order to be considered as highly variable.
HVF: Optional RNA feature names. ATAC features are selected independently.
do_scaling: Force scaling via ScaleData.
vars_to_regress: A vector of variable names to include as additional regression variables.
regression_model: "linear", "poisson", or "negativebinomial".
scale_within_batch: Scale within each batch. Only used by "Uncorrected", "Seurat", "MNN", "Harmony", "BBKNN", "CSS", "ComBat".
linear_reduction: RNA reduction, which must be "pca". ATAC uses SVD.
linear_reduction_dims: Number of components to compute for each modality.
linear_reduction_dims_use: Explicit component indices, or NULL for modality-specific selection. Inspect WNN_dims_list for the selected indices.
linear_reduction_params: Named arguments for linear reduction.
force_linear_reduction: Whether to recompute existing linear reductions.
nonlinear_reduction: One or more of "umap", "umap-naive" and "fr", or character() for graphs and clusters without an embedding.
nonlinear_reduction_dims: Number(s) of nonlinear dimensions to compute.
nonlinear_reduction_params: Shared parameters, optionally overridden by lists named after individual nonlinear reduction methods.
force_nonlinear_reduction: Recompute nonlinear reductions even when stored results exist.
neighbor_metric, neighbor_k: Neighbor graph ("euclidean", "cosine", "manhattan", "hamming").
cluster_algorithm, cluster_resolution: Clustering ("louvain", "slm", "leiden"). Larger cluster_resolution yields fewer clusters.
verbose: Whether to print the message. Default is TRUE.
seed: Random seed.

Value

A Seurat object with RNA PCA and ATAC LSI reductions, WNN neighbors, WNNKNN and WNNSNN graphs, WNNclusters metadata and requested WNN embeddings. Misc(object, "WNN_reduction_list") names the retained source reductions; WNN_dims_list records their selected dimensions. Both return modes retain complete processed modality assays (including scaled RNA), ATAC reductions, metadata and representation receipts, with the processed RNA assay (SCT when selected) active. The caller's random-number state is restored.

Details

Cells must represent paired measurements. RNA and ATAC preprocessing is followed by multimodal neighbor construction. This function does not use batch labels to correct either modality. Specify assays explicitly after adding derived assays, for example assay = c("RNA", "peaks"). Both assays must contain the same cells as the object. Subset to verified paired cells explicitly before running WNN. Nonlinear reductions supported here are "umap", "umap-naive" and "fr"; use character() to retain graphs and clusters without computing an embedding. RNA uses the requested normalization and PCA. ATAC always uses TF-IDF and SVD, including when RNA uses SCT. HVF supplies RNA features; ATAC features are selected independently from accessibility counts.

Examples

# Small paired subset; no external fragments or reference downloads.
data("pbmcmultiome_sub", package = "scop")
paired <- subset(pbmcmultiome_sub, cells = colnames(pbmcmultiome_sub)[1:80])
paired <- WNN_integrate(
  srt_merge = paired, assay = c("RNA", "peaks"),
  linear_reduction_dims = 10, linear_reduction_dims_use = 2:8,
  nHVF = 100, neighbor_k = 10, nonlinear_reduction_dims = 2,
  verbose = FALSE
)
#> ℹ [2026-11-12 06:12:59] Skip `log1p()` because `layer = data` is not "counts"
#> ℹ [2026-11-12 06:13:08] Skip `log1p()` because `layer = data` is not "counts"
#> ℹ [2026-11-12 06:13:17] Reorder clusters...
#> ℹ [2026-11-12 06:13:17] Skip `log1p()` because `layer = data` is not "counts"
SeuratObject::Reductions(paired)
#> [1] "RNApca"        "RNApcaUMAP2D"  "RNAUMAP2D"     "ATACsvd"      
#> [5] "ATACsvdUMAP2D" "ATACUMAP2D"    "ATAClsi"       "WNNUMAP2D"    
stopifnot(all(paired@misc$WNN_reduction_list %in% SeuratObject::Reductions(paired)))

Usage

Arguments

Value

Details

See also

Examples