Package 'scooter'

Description

Detected by gene symbol prefix, case-insensitively so the same patterns match both human (⁠MT-⁠, RPS/RPL, HBA/HBB) and mouse (⁠mt-⁠, Rps/Rpl, Hba-a1/Hbb-bs) naming. ⁠RPS6K*⁠ (ribosomal protein S6 kinase) is excluded from pct_ribo - it matches the RPS prefix but is a signaling kinase, not a ribosomal protein.

Usage

add_gene_class_percent(x)

Arguments

Details

High expression levels of mitochondrial genes could be an indicator of poor sample quality, leading to a high fraction of apoptotic or lysing cells (see reference below).

pct_ribo is normally 15-45%; the exact fraction varies by cell type and overall cell health. "Although both ribosomal proteins and ribosomal RNA (rRNA) make up the ribosome complex, ribosomal protein transcripts are not equivalent to ribosomal RNA (rRNA). Ribosomal protein transcript detection will not necessarily correlate with either rRNA or mitochondrial transcripts." (see reference below)

High pct_hb is a contamination signal in PBMC and most solid tissue: per-sample pct_hb flags incomplete red blood cell lysis in PBMC preps, or inadequate perfusion in solid tissue.

Value

Seurat object with pct_mito, pct_ribo, and pct_hb metadata columns added.

References

pct_mito: https://kb.10xgenomics.com/s/article/360001086611-Why-do-I-see-a-high-level-of-mitochondrial-gene-expression

pct_ribo: https://kb.10xgenomics.com/s/article/218169723-What-fraction-of-reads-map-to-ribosomal-proteins

Description

Add assay to Seurat object.

Usage

add_seurat_assay(x, assay, counts_matrix, log_file = NULL)

Arguments

Value

Seurat object of cells found in both the existing object and new data.

Description

Annotates query with Pan-human Azimuth, a neural network trained on 23 human tissues. Install the AzimuthAPI package from GitHub (satijalab/AzimuthAPI).

Usage

annotate_pan_human_azimuth(query, query_assay = "RNA", verbose = TRUE)

Arguments

Details

Unlike transfer_labels(), the function needs no reference object.

Value

query with these columns added to its metadata: pan_human_azimuth_label (the most specific label), pan_human_azimuth_final_level_confidence (its confidence), pan_human_azimuth_broad, pan_human_azimuth_medium, pan_human_azimuth_fine, pan_human_azimuth_final_level_labels, pan_human_azimuth_full_consistent_hierarchy and pan_human_azimuth_full_hierarchical_labels. The function writes the columns of AzimuthAPI, under their own names and after a cell column, to annotation/annotation-pan-human-azimuth.csv.gz. It also writes ⁠dr-umap-<column>.png⁠ for each metadata column, except those with too many groups to color.

References

Pan-human Azimuth website: https://satijalab.org/pan_human_azimuth

Sarkar S, Li Z, Molla G, Shenoy A, Zhang B, Collins D, et al. Organism-scale annotation with Pan-human Azimuth. bioRxiv. 2026 Jul 21. doi:10.64898/2026.07.16.738997

Description

Function to extract data from Seurat object.

Usage

as_data_frame_seurat(
  x,
  assay = NULL,
  slot = NULL,
  features = NULL,
  reduction = NULL,
  metadata = TRUE
)

Arguments

Value

A metadata file merged on cell identifiers.

Description

Writes ⁠expression-mean-<label>.csv⁠, and additionally ⁠expression-mean-<label>-per-sample.csv⁠ when x has more than one sample.

Usage

calculate_cluster_expression(x, label)

Arguments

Value

x, invisibly. Called for its file output.

Description

test, forwarded to Seurat::FindAllMarkers()/Seurat::FindMarkers(): "wilcox" (Wilcoxon rank sum, the default), "roc" (classification power from 0 = random to 1 = perfect), "bimod" (McDavid et al. 2013), "tobit" (Trapnell et al. 2014), or "MAST" (Finak et al. 2015).

Usage

calculate_cluster_markers(x, label, test, pairwise = FALSE)

Arguments

Details

pairwise = TRUE compares each cluster against each other cluster individually (rather than each cluster against all others combined), keeping only genes significant in every comparison - local and global markers both, at the cost of being much slower for many clusters.

Value

x, invisibly. Called for its file and plot output.

Description

Writes ⁠metadata-<label>.csv.gz⁠ (cell metadata joined with the tSNE/UMAP embeddings, skipped if neither reduction is present) and ⁠summary-<label>.csv⁠ (cell counts and percentages per cluster, split further by sample into ⁠summary-<label>-per-sample.csv⁠ when that differs).

Usage

calculate_cluster_stats(x, label)

Arguments

Value

x, invisibly. Called for its file output.

Description

Builds a shared nearest neighbor (SNN) graph from reduced dimensions and identifies clusters over a range of resolutions.

Usage

calculate_clusters(x, ...)

## S3 method for class 'Seurat'
calculate_clusters(
  x,
  assay,
  reduction,
  num_dim,
  num_neighbors = 20,
  res = NULL,
  algorithm = 4,
  log_file = NULL,
  max_clusters = length(get_color_scheme("clusters")),
  ...
)

Arguments

Details

For a Seurat object, the cluster assignments are added to the object metadata as ⁠res.<resolution>⁠ columns, the labels are converted to the C01, C02, ... convention (1-based, zero-padded, and C-prefixed to block downstream numeric coercion), and resolutions that do not yield any new clusters are dropped.

Value

Cluster assignments as a data frame (default method) or the Seurat object with the cluster assignments added to its metadata (Seurat method).

Description

Check identity of the Seurat object.

Usage

check_identity_column(x, identity_column)

Arguments

Value

The name of the identity column, potentially corrected if resolution.

Description

The clustering step of the pipeline in one call: run tSNE and UMAP on an existing "pca" reduction (run_tsne(), run_umap()), build the SNN graph and cluster over a range of resolutions (calculate_clusters()), and plot each reduction by cluster.

Usage

cluster_seurat_object(
  x,
  assay = "RNA",
  reduction,
  num_dim,
  num_neighbors = 30,
  metadata_file = NULL,
  log_file = NULL
)

Arguments

Details

This function computes both reductions before clustering, since the cluster plots draw on them. run_tsne() and run_umap() already write their own sample-colored scatter to the working directory, so this function skips plotting "by sample" again.

The per-resolution cluster plots also write unconditionally, into a ⁠clusters-resolutions/⁠ subdirectory. This matches every other verb's standard-output convention.

Value

A Seurat object with "tsne" and "umap" reductions and one ⁠res.<x>⁠ column per retained clustering resolution.

Description

Function to create a color vector.

Usage

create_color_vect(seurat_obj, group = "orig.ident")

Arguments

Value

A vector of colors.

Description

Everything needed to take a sample from a counts file to an object ready for clustering, in one call.

Usage

create_seurat_object(
  path,
  sample_name,
  num_mads = 3,
  min_genes = 500,
  max_genes = NULL,
  min_counts = NULL,
  max_counts = NULL,
  max_mt = 10,
  min_cells = 100,
  normalization_method = "log",
  num_variable_genes = 3000,
  num_pcs = 50,
  num_dim = 30,
  log_file = NULL
)

Arguments

Details

The function runs these steps:

1. Read the counts (read_counts_file()).

2. Build the object (initialize_seurat_object()).

3. Attach any antibody capture data as an "ADT" assay (add_seurat_assay()).

4. Apply the quality cutoffs (filter_cells()).

5. Normalize, select variable features, and scale (normalize_counts(), which does all three for either method).

6. Run PCA (run_pca()).

7. Compute a first-pass tSNE/UMAP (run_tsne(), run_umap()).

This function attaches the ADT assay before filtering, on purpose. add_seurat_assay() restricts the object to the barcodes both matrices share. Attaching ADT after filtering would apply the cutoffs to a different set of cells.

PCA, and the tSNE/UMAP computed from it, runs on whichever assay normalization left active. SCTransform() creates and activates "SCT". The log path leaves "RNA" active. So dr_assay is decided directly from normalization_method, not read back off the object. Reading it back off DefaultAssay() would reintroduce the implicit dependency an explicit assay argument exists to avoid.

The tSNE/UMAP computed here are a first-pass look, capped at num_dim dimensions. cluster_seurat_object() recomputes both once a clustering resolution's own num_dim is chosen. So this pair is not the final one used for the actual clustering plots.

Value

A Seurat object with normalized data, variable features, scaled data, and "pca", "tsne", and "umap" reductions — ready for cluster_seurat_object(). Also writes the variable-feature, PCA, tSNE, and UMAP diagnostic plots (via normalize_counts(), run_pca(), run_tsne(), and run_umap()) and the metadata+embeddings table (via save_metadata(), as metadata.csv.gz) to the working directory.

Description

Calculate differentially expressed genes within each subpopulation/cluster

Usage

differential_expression_per_cluster(
  x,
  cluster_column,
  group_column,
  test = "wilcox",
  out_path = ".",
  write = TRUE,
  log_file = NULL
)

Arguments

Value

.

Description

A label-count estimate made without building the plot. plot_dr_group() now sizes itself from the actual rendered legend instead (see save_dr_plot()), so this has no callers left in the package - kept as a cheap standalone estimate for callers that don't have a plot object yet.

Usage

dr_plot_width(x, group_by = "orig.ident")

Arguments

Value

Numeric width in inches.

Description

Filtering runs in two stages, and the log records how many cells each one removed.

Usage

filter_cells(
  x,
  num_mads = 3,
  min_genes = 500,
  max_genes = NULL,
  min_counts = NULL,
  max_counts = NULL,
  max_mt = 10,
  min_cells = 50,
  log_file = NULL
)

Arguments

Details

1. Outlier removal. Cells further than num_mads median absolute deviations from the median number of genes or counts are dropped. The median and MAD are computed on the log scale and the bounds back-transformed, which suits these right-skewed counts. num_mads = NULL skips it.

2. Fixed cutoffs. min_genes / max_genes / min_counts / max_counts are applied to whatever survived stage 1, then max_mt. Each is skipped when NULL, so out of the box only min_genes (500) and max_mt (10) do anything.

Applying the fixed cutoffs second makes them a quality floor the data-driven bounds cannot undercut: a wide or contaminated distribution can put the MAD bound below 500 genes, and those cells are still removed. All cutoffs are inclusive.

Genes that are no longer detected in any of the retained cells are dropped, since filtering cells leaves them as all-zero rows.

Value

Filtered data. The Seurat method also writes a pre-filtering QC violin plot (via plot_metrics_distribution()) and the unfiltered metadata (via save_metadata(), as metadata-unfiltered.csv.gz), plus the post-filtering QC plots (via plot_metrics_distribution() and plot_metrics_correlations()) to the working directory.

Description

Get geneset scores.

Usage

geneset_score(module_tbl, counts_raw, min_cpm = 0, limit_pct = 1)

Arguments

Value

.

Description

Determine the color scheme. Can be specified for samples or clusters to avoid confusion.

Usage

get_color_scheme(type = "clusters")

Arguments

Value

A vector of colors.

Description

Determine the point size for reduced dimensions scatter plots (smaller for larger datasets).

Usage

get_dr_point_size(x)

Arguments

Value

Numeric point size.

Description

Get a small matrix of raw counts from the PBMC dataset.

Usage

get_test_counts_matrix()

Value

A matrix of raw counts.

Examples

pbmc_mat <- get_test_counts_matrix()

Description

Read in 10x Genomics Cell Ranger Matrix Market format data.

Usage

import_mtx(data_path, gene_column = 2, prefix = "", log_file = NULL)

Arguments

Value

Named list of matrices. One matrix for each data type as specified in the third column of the features.tsv file. As of Oct 3rd 2019, the two options are ⁠Gene Expression⁠ and ⁠Antibody Capture⁠

Description

For an RNA assay the per-cell QC metrics are added at the same time: pct_mito/pct_ribo/ pct_hb (via add_gene_class_percent()), plus detected_genes and total_counts as clearer aliases of nFeature_RNA/nCount_RNA, and sample_name as a clearer alias of orig.ident. All six are metadata columns as soon as the object exists, so plot_metrics_distribution() (which takes its metrics explicitly, no default) and plot_metrics_correlations() can use any of them without further preparation.

Usage

initialize_seurat_object(
  counts_matrix,
  assay = "RNA",
  min_cells = 1,
  min_genes = 1,
  project = "proj",
  log_file = NULL
)

Arguments

Value

Seurat object.

Description

Runs the Seurat 5 IntegrateLayers() workflow. It adds what calling IntegrateLayers() directly leaves for the caller to handle:

Usage

integrate_layers(
  x,
  assay = "RNA",
  int_reduction = "cca",
  num_dim,
  k_anchor = 10,
  k_weight = 100,
  log_file = NULL,
  ...
)

Arguments

Details

• Clamps k_weight to fit the smallest batch. Seurat's own default of 100 errors outright below that size. See k_weight below.

• Excludes dims/k.anchor for "harmony". HarmonyIntegration() silently ignores both instead of erroring, so passing them would rely on that silent drop. See num_dim/k_anchor below.

• Forces future::plan(future::sequential). Benchmarks show multisession gives no gain here, and is a documented liability at this scale elsewhere in this package.

• Rejoins the split layers with JoinLayers() afterward. IntegrateLayers() itself leaves this undone.

Unlike the older anchor workflow, this function creates no "integrated" assay. Clustering and visualization use the corrected embedding directly, instead of re-running PCA. This function expects split_layers_by_batch() to have run first. It is the one step of integrate_seurat_object() that calls IntegrateLayers() directly.

Value

A Seurat object with rejoined layers and an int_reduction-named reduction ("cca", "rpca", or "harmony").

Description

The integrate step of the pipeline in one call. It plays the same role for a merged object that create_seurat_object() plays for a single sample.

Usage

integrate_seurat_object(
  x,
  assay = "RNA",
  int_reduction = "cca",
  num_dim,
  batch_var = "orig.ident",
  num_variable_genes = 3000,
  num_neighbors = 30,
  k_anchor = 10,
  k_weight = 100,
  log_file = NULL,
  ...
)

Arguments

Details

The function runs these steps:

1. Split the assay layers by batch and prepare them (split_layers_by_batch()).

2. Report how much the per-batch variable genes overlap (plot_var_genes_euler(), plot_var_genes_upset()).

3. Correct the batch effect (integrate_layers()).

4. Compute a first-pass tSNE/UMAP on the corrected embedding.

The tSNE/UMAP from step 4 is a preview, capped at num_dim. It differs from the real one cluster_seurat_object() computes later over its own num_dim. cluster_seurat_object() has no default reduction. Pass it this function's int_reduction name explicitly. Otherwise it may use the uncorrected "pca" reduction that split_layers_by_batch() also leaves on the object.

Value

A Seurat object with rejoined layers, an int_reduction-named reduction ("cca", "rpca", or "harmony"), and "tsne"/"umap" reductions computed from it. Also writes the variable-gene overlap, QC distribution, integrated-embedding, and tSNE/UMAP plots to the working directory.

Description

x and y must not share column names other than cell - stop()s if they do, rather than letting left_join() silently rename the collision to ⁠<col>.x⁠/⁠<col>.y⁠.

Usage

merge_metadata(x, y, log_file = NULL)

Arguments

Value

A metadata file merged on cell identifiers.

Description

Merges a list of Seurat objects, rejoins the split layers that Seurat 5 leaves behind, and removes genes that are detected in too few cells. Features of the non-RNA assays (such as ADT and HTO) are always kept.

Usage

merge_seurat_objects(seurat_object_list, min_cells = NULL, log_file = NULL)

Arguments

Details

Elements of seurat_object_list may be Seurat objects, or paths to an analysis directory containing a seurat_obj.qs2 (as written by the CLI). Which one it is does not matter to the caller: this function resolves each element to an object first.

Once every input is an object, this function strips each one's own reductions, scaled data, and clustering columns before merging. A per-sample PCA, tSNE/UMAP, or clustering — written by create_seurat_object()/cluster_seurat_object() — is meaningless once merged. normalize_counts()/run_pca() recompute all of it downstream anyway.

Writes metrics-distribution.png, a per-sample QC violin plot, to the working directory.

Value

A merged Seurat object.

Description

A thin wrapper that picks the right Seurat function for method and fixes the parameters this pipeline uses, so either path hands back a fully processed object — no separate variance step needed. method = "sct" already worked this way natively: SCTransform() produces the counts, variable features and scaled data in one pass. method = "log" now mirrors that explicitly: NormalizeData() followed by FindVariableFeatures() and ScaleData().

Usage

normalize_counts(
  x,
  method = "log",
  num_variable_genes = 3000,
  assay = "RNA",
  log_file = NULL
)

Arguments

Details

No covariates are regressed out (vars.to.regress = NULL) for either path. Germain et al. (2020) found that the "common practice of regressing out cell covariates, such as the detection rate or proportion of mitochondrial reads[,] nearly always had a negative impact."

On the log path, this function drops any ⁠vf_*⁠ columns already on assay's feature meta data before recomputing them. An earlier call can leave these columns behind: each sample's own pre-merge run, or a merge step's preview run. Left alone, they would just accumulate call after call. See variable_features_by_batch() for what reading an accumulated column list, instead of a current one, leads to.

SCTransform() never writes this kind of per-gene bookkeeping column. So the sct path has nothing to clean up. The check runs the same way for both paths; it is a no-op for sct, not something conditioned on method.

Value

The processed Seurat object. Also writes the variable-feature plot (variance-features.png) to the working directory.

Description

Heatmap of the top cluster markers, at several top-N cutoffs.

Usage

plot_cluster_markers_heatmap(x, markers_tbl, num_genes, filename_base)

Arguments

Value

Nothing; called for its file output.

Description

Plot a gene list in multiple ways: UMAP, dot plot, violin, and per-cluster bar plot.

Usage

plot_cluster_markers_top(
  x,
  genes,
  filename_base,
  color_scheme = get_color_scheme("clusters")
)

Arguments

Value

Nothing; called for its file output.

Description

A single cluster says nothing, and past the color scheme's length the colors start repeating, so nothing is plotted outside 2:length(color_scheme) clusters.

Usage

plot_clusters(
  x,
  resolution,
  filename_base,
  color_scheme = get_color_scheme("clusters")
)

Arguments

Value

x with resolution set as its identity.

Description

Plot the distribution of specified features/variables.

Usage

plot_distribution(x, features, grouping, color_scheme = NULL)

Arguments

Value

A vector of colors.

Description

One point per cell, colored by expression/abundance of feature (a gene, antibody, or any other value FeaturePlot() can fetch). Unlike passing a multi-color ramp straight to FeaturePlot(cols = ), the ramp is applied afterward with scale_color_gradientn() so the color scale stays continuous - see the "cols" note below.

Usage

plot_dr_feature(
  x,
  feature,
  assay = NULL,
  reduction = "umap",
  cells = NULL,
  color_scheme = NULL,
  pt_size = NULL,
  file_prefix = NULL,
  file_format = c("png", "pdf")
)

Arguments

Value

A ggplot object.

Seurat's FeaturePlot(cols = ) quirk

As of Seurat 5, passing a cols vector of anything other than exactly 2 colors makes FeaturePlot() bin the continuous values into 2 groups before coloring (cut(x, breaks = 2)), which turns a smooth expression gradient into a flat low/high split. Seurat 4 only did this when cols had exactly 2 colors, so a long custom ramp used to pass through untouched. This function sidesteps the quirk (and the version difference) entirely by never passing a multi-color cols to FeaturePlot() - the ramp is layered on afterward instead.

Description

The categorical counterpart of plot_dr_feature(): cells are colored by a metadata column (or by the current identities) rather than by a continuous feature. Cells are shuffled so no one group is drawn on top of the others.

Usage

plot_dr_group(
  x,
  reduction = "umap",
  group_by = NULL,
  color_scheme = NULL,
  pt_size = NULL,
  na_value = "grey50",
  file_prefix = NULL,
  file_format = c("png", "pdf"),
  width = NULL,
  height = 6
)

Arguments

Value

A ggplot object.

Description

Plot a UMAP colored by one or more cluster/grouping variables.

Usage

plot_dr_umap_clusters(
  x,
  cluster_var,
  color_scheme = get_color_scheme("clusters")
)

Arguments

Value

Nothing; called for its file output, one ⁠dr-umap-<cluster_var>.png⁠ per variable.

Description

One panel per pair in pairs, colored by sample. No legend, since this only ever runs on a single sample - see filter_cells(). Pairs are a curated list rather than every combination of a metric vector, since that would make the panel count quadratic in the number of metrics (5 metrics -> 10 panels via combn()) and most pairs are not biologically meaningful together.

Usage

plot_metrics_correlations(
  x,
  pairs = list(c("total_counts", "detected_genes"), c("pct_mito", "detected_genes"),
    c("pct_mito", "pct_ribo")),
  group_by = "orig.ident",
  color_scheme = NULL,
  file = NULL,
  width = 16,
  height = 5
)

Arguments

Value

A combined plot, one panel per pair.

Description

One violin panel per metric, grouped by sample. The metrics are metadata columns, so layer is set explicitly: VlnPlot() otherwise searches for the "data" layer, which does not exist before NormalizeData() has been run.

Usage

plot_metrics_distribution(
  x,
  metrics,
  group_by = "orig.ident",
  color_scheme = NULL,
  file = NULL,
  width = 12,
  height = 6
)

Arguments

Value

A combined plot, one panel per plotted metric.

Description

Becomes unreadable and can take a very long time to fit for many sets, so returns NULL when there are too many batches.

Usage

plot_var_genes_euler(var_genes_list, color_scheme = NULL, max_sets = 8)

Arguments

Value

A plot, or NULL if there are too many sets to draw.

Description

UpSet plot of the variable genes shared between batches.

Usage

plot_var_genes_upset(var_genes_list, nsets = 50, nintersects = 15)

Arguments

Value

An UpSet plot.

Description

Works on any S4 object or list, not just Seurat objects. It descends into S4 slots and list elements, bounded by max_depth. So a Seurat object, an assay pulled from one (x[["RNA"]]), a DimReduc, or a plain nested list all get the same recursive breakdown.

Usage

profile_object_size(x, max_depth = 3)

Arguments

Details

An Assay5's layers slot is itself a list. So layers > counts/layers > data/layers > scale.data are sized separately, not lumped into one layers total.

A matrix, data frame, or atomic vector is neither S4 nor a list. Passing one of these just reports its own size, the same as calling object.size() directly. There is nothing to descend into.

Value

A tibble with one row per slot/element, largest first: path (its position, " > "-separated) and size (human-readable, e.g. "1.2 Mb").

Description

Read in Gene Expression and Antibody Capture data from a 10x Genomics Cell Ranger sparse matrix or from a text file.

Usage

read_counts_file(path, sample_name, log_file = NULL)

Arguments

Value

Named list of matrices. One matrix for each data type. Currently the only two data types are 'Gene Expression' and 'Antibody Capture'

Description

Meant for parameters that accept either an in-memory Seurat object or a serialized one - resolving here means callers do not need to check which they got themselves.

Usage

resolve_seurat_object(x)

Arguments

Value

A Seurat object.

Description

Run dimensionality reduction: pca, tsne, or umap

Usage

run_dr(x, dr_method, suffix = "", assay = "RNA", ...)

Arguments

Details

run_dr() is a thin dispatcher: it routes to run_pca(), run_tsne(), or run_umap() based on dr_method, forwarding suffix and anything in .... It only carries the parameters common to every method; method-specific ones (num_pcs, num_neighbors, reduction, ...) are passed through ... to the chosen function. Each of those functions is itself generic: a plain matrix takes the .default method (irlba/Rtsne/uwot), a Seurat object takes the .Seurat method (native RunPCA()/RunTSNE()/RunUMAP(), storing a "<method><suffix>" reduction).

Value

The reduced data (matrix input) or a Seurat object with the reduction added. For a Seurat object, also writes the diagnostic plot(s) for the chosen method to the working directory: PCA gets a sample-colored scatter and an elbow plot (plus a heatmap past 15 components); tSNE/UMAP get a sample-colored scatter.

See Also

run_pca(), run_tsne(), run_umap()

Description

The default method runs irlba::prcomp_irlba() on a matrix; the Seurat method runs Seurat::RunPCA() on the object and stores a "pca<suffix>" reduction.

Usage

run_pca(x, ...)

## Default S3 method:
run_pca(x, num_pcs = 50, suffix = "", ...)

## S3 method for class 'Seurat'
run_pca(
  x,
  num_pcs = 50,
  suffix = "",
  assay = "RNA",
  features = NULL,
  var_features = FALSE,
  ...
)

Arguments

Value

A named list (default) or a Seurat object with the reduction added (Seurat method). The Seurat method also writes the diagnostic plots to the working directory: a sample-colored scatter and an elbow plot, plus a heatmap past 15 components.

Description

The default method runs Rtsne::Rtsne() on a matrix; the Seurat method runs Seurat::RunTSNE() on the object and stores a "tsne<suffix>" reduction.

Usage

run_tsne(x, ...)

## Default S3 method:
run_tsne(x, seed.use = 1, dim.embed = 2, suffix = "", ...)

## S3 method for class 'Seurat'
run_tsne(
  x,
  seed.use = 1,
  dim.embed = 2,
  suffix = "",
  assay = "RNA",
  reduction = NULL,
  num_dim = NULL,
  features = NULL,
  var_features = FALSE,
  file_format = c("png", "pdf"),
  ...
)

Arguments

Value

A tSNE embedding matrix (default) or a Seurat object with the reduction added. The Seurat method also writes a sample-colored scatter of the result to the working directory, as ⁠dr.<reduction>.png⁠/.pdf (or just .png if file_format = "png").

Description

The default method runs uwot::umap2() on a matrix; the Seurat method runs Seurat::RunUMAP() with umap.method = "uwot2" (also uwot::umap2()) and stores a "umap<suffix>" reduction. Defaults match Seurat::RunUMAP() (30 neighbors, minimum distance 0.3) rather than the lower uwot defaults.

Usage

run_umap(x, ...)

## Default S3 method:
run_umap(x, num_neighbors = 30, min_dist = 0.3, suffix = "", ...)

## S3 method for class 'Seurat'
run_umap(
  x,
  num_neighbors = 30,
  min_dist = 0.3,
  suffix = "",
  assay = "RNA",
  reduction = NULL,
  num_dim = NULL,
  features = NULL,
  var_features = FALSE,
  graph = NULL,
  file_format = c("png", "pdf"),
  ...
)

Arguments

Value

A UMAP embedding matrix (default) or a Seurat object with the reduction added. The Seurat method also writes a sample-colored scatter of the result to the working directory, as ⁠dr.<reduction>.png⁠/.pdf (or just .png if file_format = "png").

Description

Layers left split by a merge() are rejoined first, since LayerData() would otherwise return only the first of them, and warn rather than error while doing it.

Usage

save_counts(x, assay = "RNA", layer = "data", digits = 3, file)

Arguments

Value

A tibble with one row per gene.

Description

Unlike dr_plot_width() (a label-count guess made before the plot exists), this measures the legend actually rendered on plot - so long label text widens the saved image even when the label count is small. Requires an open graphics device to get accurate text metrics, so it opens and closes a throwaway one around the measurement.

Usage

save_dr_plot(
  plot,
  file_prefix,
  file_format = c("png", "pdf"),
  width = NULL,
  height = 6,
  panel_width = height
)

Arguments

Value

plot, invisibly.

Description

Joins the requested reductions onto the cell metadata on a cell column. Reductions that are not present in the object are skipped, so this works before and after the dimensionality reduction steps have run.

Usage

save_metadata(
  x,
  reduction = c("tsne", "umap"),
  digits = 3,
  file = "metadata.csv.gz"
)

Arguments

Value

The table written to file, invisibly.

Description

Wrapper for SeuratObject::Idents() with extra safety checks.

Usage

set_identity(x, identity_column)

Arguments

Value

A Seurat object with an updated identity set.

Description

Drops the scaled data, reductions, graphs, and clustering columns, then splits the assay layers by a metadata column — splitting the layers rather than the object is what the Seurat 5 IntegrateLayers() workflow expects.

Usage

split_layers_by_batch(
  x,
  batch_var = "orig.ident",
  assay = "RNA",
  num_variable_genes = 3000,
  num_dim = 50,
  log_file = NULL
)

Arguments

Details

Everything after the split runs through the same verbs as the single-sample path (normalize_counts(), run_pca()), so the two workflows cannot drift apart on parameters. integrate_layers() then consumes the resulting "pca" reduction.

Value

A Seurat object with split layers and a "pca" reduction.

Description

Reference-based annotation wrapping Seurat's reference-mapping workflow (Seurat::FindTransferAnchors() + Seurat::TransferData()) to annotate query with labels (cell type, or any other categorical metadata) from an already integrated and annotated reference object. Optionally also projects query onto the reference's UMAP via Seurat::MapQuery().

Usage

transfer_labels(
  query,
  ref,
  query_label_col,
  ref_label_col,
  query_assay,
  ref_assay,
  normalization,
  ref_reduction = "pca",
  num_dim,
  anchor_reduction = "pcaproject",
  ref_umap = FALSE,
  verbose = TRUE
)

Arguments

Value

query with query_label_col (a factor) added to its metadata, and a "ref.umap" reduction added if ref_umap = TRUE. Also writes the predictions table (cell/predicted.id/prediction.score.max) to ⁠annotation/annotation-transfer-label-<query_label_col>.csv.gz⁠, and a ⁠dr-<reduction>-<query_label_col>.png⁠ UMAP colored by the transferred labels for each of "umap"/"ref.umap" already present on query.

Description

After split_layers_by_batch() the per-batch variable features are stored as ⁠vf_<method>_<layer>_variable⁠ columns in the assay meta data rather than being reachable through VariableFeatures(), which returns the combined set.

Usage

variable_features_by_batch(x, assay = "RNA")

Arguments

Details

normalize_counts() now clears its own assay's ⁠vf_*⁠ columns before recomputing them. But an object built before that guard existed can still carry a whole history of them: each sample's own pre-merge run from create_seurat_object(), and the merge step's own post-merge preview run. DietSeurat() clears layers, reductions, and graphs, but never assay meta data. merge() disambiguates colliding names with a numeric suffix, for example vf_vst_counts.1_variable.

Matching every ⁠_variable$⁠ column would pick up that whole history along with the batches actually being asked about. This function instead anchors on Layers(x[[assay]], search = "counts"), the object's current split state. That keeps stale columns out, regardless of which generation of the object this is.

Value

A named list of variable gene vectors, one element per batch.

Description

Small function to write to message and to log file.

Usage

write_message(message_str, log_file = NULL)

Arguments

Value

A message and writes the message to the specified log file.

Examples

write_message(message_str = "Finished Step 1", log_file = "log.file.txt")