plotScoreHeatmap: Plot a score heatmap

In SingleR: Reference-Based Single-Cell RNA-Seq Annotation

Description

Create a heatmap of the SingleR assignment scores across all cell-label combinations.

Usage

plotScoreHeatmap(
  results,
  cells.use = NULL,
  labels.use = NULL,
  clusters = NULL,
  show.labels = TRUE,
  show.pruned = FALSE,
  max.labels = 40,
  normalize = TRUE,
  cells.order = NULL,
  order.by = c("labels", "clusters"),
  scores.use = NULL,
  calls.use = 0,
  na.color = "gray30",
  cluster_cols = FALSE,
  annotation_col = NULL,
  show_colnames = FALSE,
  color = (grDevices::colorRampPalette(c("#D1147E", "white", "#00A44B")))(100),
  silent = FALSE,
  ...,
  grid.vars = list()
)

Arguments

results	A DataFrame containing the output from SingleR, classifySingleR, combineCommonResults, or combineRecomputedResults.
cells.use	Integer or string vector specifying the single cells (i.e., rows of results) to show. If NULL, all cells are shown.
labels.use	Character vector specifying the labels to show in the heatmap rows. Defaults to all labels in results.
clusters	String vector or factor containing cell cluster assignments, to be shown as an annotation bar in the heatmap.
show.labels	Logical indicating whether the assigned labels should be shown as an annotation bar.
show.pruned	Logical indicating whether the pruning status of the cell labels, as defined by pruneScores, should be shown as an annotation bar.
max.labels	Integer scalar specifying the maximum number of labels to show.
normalize	Logical specifying whether correlations should be normalized to lie in [0, 1].
cells.order	Integer or String vector specifying how to order the cells/columns of the heatmap. Regardless of cells.use, this input should be the the same length as the total number of cells. Ignored if cluster_cols is set.
order.by	String providing the annotation to be used for cells/columns ordering. Can be "labels" (default) or "clusters" (when provided). Ignored if cells.order or cluster_cols are specified.
scores.use	Integer scalar or vector specifying the individual annotation result from which to take scores. This is only relevant for combined results, see Details.
calls.use	Integer scalar or vector specifying the individual annotation result from which to take labels, for use in the annotation bar when show.labels=TRUE. This is only relevant for combined results, see Details.
na.color	String specifying the color for non-calculated scores of combined results.
annotation_col, cluster_cols, show_colnames, color, silent, ...	Additional parameters for heatmap control passed to pheatmap.
grid.vars	A named list of extra variables to pass to grid.arrange, used to arrange the multiple plots generated when scores.use is of length greater than 1.

Details

This function creates a heatmap containing the SingleR initial assignment scores for each cell (columns) to each reference label (rows). Users can then easily identify the high-scoring labels associated with each cell and/or cluster of cells.

If show.labels=TRUE, an annotation bar will be added to the heatmap showing the label assigned to each cell. This is also used to order the columns for a more organized visualization when order.by="label". Note that scores shown in the heatmap are initial scores prior to the fine-tuning step, so the reported labels may not match up to the visual maximum for each cell in the heatmap.

If max.labels is less than the total number of unique labels, only the top labels are shown in the plot. Labels that were called most frequently are prioritized. The remaining labels are then selected based on:

• Labels with max z-scores after per-cell centering and scaling of the scores matrix, if results does not contain combined scores.

• Labels which were suggested most frequently by individual references, if results contains combined scores.

Value

If scores.use specifies a single set of scores, the output of pheatmap is returned showing the heatmap on the current graphics device.

If scores.use specifies multiple scores for a combined result, multiple heatmaps are generated in a grid on the current graphics device.

If scores.use specifies multiple scores and grid.vars is set to NULL, a list is returned containing the pheatmap globs for manual display.

Working with combined results

For combined results (see ?combineRecomputedResults), this function can show both the combined and individual scores or labels. This is done using the scores.use and calls.use arguments, entries of which refer to columns of results$orig.results if positive or to the combined results if zero. For example:

• If we set scores.use=2 and calls.use=1, we will plot the scores from the second individual reference with the annotation bar containing label assignments from the first reference.

• If we set scores.use=1:2 and calls.use=1:2, we will plot the scores from first and second references (in separate plots) with the annotation bar in each plot containing the corresponding label assignments.

• By default, the function will create a separate plot the combined scores and each individual reference. In each plot, the annotation bar contains the combined labels; this is equivalent to scores.use=0:N and calls.use=0 for N individual references.

Tweaking the output

Additional arguments can be passed to pheatmap for further tweaking of the heatmap. Particularly useful parameters are show_colnames, which can be used to display cell/cluster names; treeheight_row, which sets the width of the clustering tree; and annotation_col, which can be used to add extra annotation layers. Clustering, pruning and label annotations are automatically generated and appended to annotation_col when available.

Normalization of colors

If normalize=TRUE, scores will be linearly adjusted for each cell so that the smallest score is 0 and the largest score is 1. This is followed by cubing of the adjusted scores to improve dynamic range near 1. Visually, the color scheme is changed to a blue-green-yellow scale.

The adjustment is intended to inflate differences between scores within a given cell for easier visualization. This is because the scores are often systematically shifted between cells, making the raw values difficult to directly compare. However, it may be somewhat misleading; fine-tuning may appear to assign a cell to a label with much lower score whereas the actual scores are much closer. It is for this reason that the color bar values are not shown as the absolute values of the score have little meaning.

Note that this transformation is not dependent on the choice of the top max.labels labels. Altering max.labels will not change the normalized values, only the labels that are shown. However, the transformation will respond to labels.use.

Author(s)

Daniel Bunis, based on code by Dvir Aran.

See Also

SingleR, to generate scores.

pruneScores, to remove low-quality labels based on the scores.

pheatmap, for additional tweaks to the heatmap.

grid.arrange, for tweaks to the how heatmaps are arranged when multiple are output together.

Examples

# Running the SingleR() example.
example(SingleR, echo=FALSE)

# Grab the original identities of the cells as mock clusters
clusts <- test$label

# Creating a heatmap with just the labels.
plotScoreHeatmap(pred)

# Creating a heatmap with clusters also displayed.
plotScoreHeatmap(pred,
    clusters=clusts)

# Creating a heatmap with whether cells were pruned displayed.
plotScoreHeatmap(pred,
    show.pruned = TRUE)

# We can also turn off the normalization with Normalize = FALSE
plotScoreHeatmap(pred, clusters=clusts,
    normalize = FALSE)

# To only show certain labels, you can use labels.use or max.labels
plotScoreHeatmap(pred, clusters=clusts,
    labels.use = c("A","B","D"))
plotScoreHeatmap(pred, clusters=clusts,
    max.labels = 4)

# We can pass extra tweaks the heatmap as well
plotScoreHeatmap(pred, clusters=clusts,
    fontsize_row = 20)
plotScoreHeatmap(pred, clusters=clusts,
    treeheight_row = 15)
plotScoreHeatmap(pred, clusters=clusts, cluster_col = TRUE,
    cutree_cols = 5)

### Multi-Reference Compatibility ###

example(combineRecomputedResults, echo = FALSE)
plotScoreHeatmap(combined)

# 'scores.use' sets which particular run's scores to show, and can be multiple
plotScoreHeatmap(combined,
    scores.use = 1)
plotScoreHeatmap(combined,
    scores.use = c(0,2))

# 'calls.use' adjusts which run's labels and pruning calls to display.
plotScoreHeatmap(combined,
    calls.use = 1)

# To have plots output in a grid rather than as separate pages, provide,
# a list of inputs for gridExtra::grid.arrange() to 'grids.vars'.
plotScoreHeatmap(combined,
    grid.vars = list(ncol = 1))

# An empty list will use grid.arrange defaluts
plotScoreHeatmap(combined,
    grid.vars = list())

SingleR documentation built on Feb. 4, 2021, 2:01 a.m.