plotExprHeatmap: Plot expression heatmap
In CATALYST: Cytometry dATa anALYSis Tools

Description Usage Arguments Details Value Author(s) References See Also Examples

Heatmap of marker expressions aggregated by sample, cluster, or both; with options to include annotation of cell metadata factors, clustering(s), as well as relative and absolute cell counts.

plotExprHeatmap(
  x,
  features = NULL,
  by = c("sample_id", "cluster_id", "both"),
  k = "meta20",
  m = NULL,
  assay = "exprs",
  fun = c("median", "mean", "sum"),
  scale = c("first", "last", "never"),
  q = 0.01,
  row_anno = TRUE,
  col_anno = TRUE,
  row_clust = TRUE,
  col_clust = TRUE,
  row_dend = TRUE,
  col_dend = TRUE,
  bars = FALSE,
  perc = FALSE,
  bin_anno = FALSE,
  hm_pal = rev(brewer.pal(11, "RdYlBu")),
  k_pal = CATALYST:::.cluster_cols,
  m_pal = k_pal,
  distance = c("euclidean", "maximum", "manhattan", "canberra", "binary", "minkowski"),
  linkage = c("average", "ward.D", "single", "complete", "mcquitty", "median",
    "centroid", "ward.D2")
)

`x`	a `SingleCellExperiment`.
`features`	character string specifying which features to include; valid values are `"type"/"state"` for `type/state_markers(x)` if `rowData(x)$marker_class` have been specified; a subset of `rownames(x)`; NULL to use all features. When `by = "both"`, only 1 feature is allowed.
`by`	character string specifying whether to aggregate by sample, cluster, both.
`k`	character string specifying which clustering to use when `by != "sample_id"`; `assay` data will be aggregated across these cluster IDs.
`m`	character string specifying a metaclustering to include as an annotation when `by != "sample_id"` and `row_anno = TRUE`.
`assay`	character string specifying which assay data to use; valid values are `assayNames(x)`.
`fun`	character string specifying the function to use as summary statistic.
`scale`	character string specifying the scaling strategy: `"first"`: scale & trim then aggregate `"last"`: aggregate then scale & trim `"never"`: aggregate only If `scale != "never"`, data will be scaled using lower (`q`%) and upper (`1-q`%) quantiles as boundaries.
`q`	single numeric in [0,0.5) determining the quantiles to trim when `scale != "never"`.
`row_anno, col_anno`	logical specifying whether to include row/column annotations (see details); when one axis corresponds to samples (`by != "cluster_id"`), this can be a character vector specifying a subset of `names(colData(x))` to be included as annotations.
`row_clust, col_clust`	logical specifying whether rows/columns should be hierarchically clustered and re-ordered accordingly.
`row_dend, col_dend`	logical specifying whether to include the row/column dendrograms.
`bars`	logical specifying whether to include a barplot of cell counts per cluster as a right-hand side row annotation.
`perc`	logical specifying whether to display percentage labels next to bars when `bars = TRUE`.
`bin_anno`	logical specifying whether to display values inside bins.
`hm_pal`	character vector of colors to interpolate for the heatmap.
`k_pal, m_pal`	character vector of colors to interpolate for cluster annotations when `by != "sample_id"`.
`distance`	character string specifying the distance metric to use for both row and column hierarchical clustering; passed to `Heatmap`
`linkage`	character string specifying the agglomeration method to use for both row and column hierarchical clustering; passed to `Heatmap`

By default (row/col_anno = TRUE), for axes corresponding to samples (y-axis for by = "sample_id" and x-axis for by = "both"), annotations will be drawn for all non-numeric cell metadata variables. Alternatively, a specific subset of annotations can be included for only a subset of variables by specifying row/col_anno to be a character vector in names(colData(x)) (see examples).

For axes corresponding to clusters (y-axis for by = "cluster_id" and "both"), annotations will be drawn for the specified clustering(s) (arguments k and m).

a Heatmap-class object.

Helena L Crowell helena.crowell@uzh.ch

Nowicka M, Krieg C, Crowell HL, Weber LM et al. CyTOF workflow: Differential discovery in high-throughput high-dimensional cytometry datasets. F1000Research 2017, 6:748 (doi: 10.12688/f1000research.11622.1)

plotMedExprs, plotFreqHeatmap, plotMultiHeatmap

data(PBMC_fs, PBMC_panel, PBMC_md)
sce <- prepData(PBMC_fs, PBMC_panel, PBMC_md)
sce <- cluster(sce) 

# median scaled & trimmed expression by cluster
plotExprHeatmap(sce, 
  by = "cluster_id", k = "meta8",
  scale = "first", q = 0.05, bars = FALSE)

# scale each marker between 0 and 1 
# after aggregation (without trimming)
plotExprHeatmap(sce, 
  scale = "last", q = 0,
  bars = TRUE, perc = TRUE,
  hm_pal = hcl.colors(10, "YlGnBu", rev = TRUE))

# raw (un-scaled) median expression by cluster-sample
plotExprHeatmap(sce,
  features = "pp38", by = "both", k = "meta10", 
  scale = "never", row_anno = FALSE, bars = FALSE)
  
# include only subset of samples
sub <- filterSCE(sce, 
  patient_id != "Patient",
  sample_id != "Ref3")
 
# includes specific annotations &
# split into CDx & all other markers
is_cd <- grepl("CD", rownames(sce))
plotExprHeatmap(sub, 
  rownames(sce)[is_cd], 
  row_anno = "condition", 
  bars = FALSE)
plotExprHeatmap(sub, 
  rownames(sce)[!is_cd], 
  row_anno = "patient_id",
  bars = FALSE)