How to Make a Heatmap • jamses

Load R packages:

library(jamses)
suppressPackageStartupMessages(
library(SummarizedExperiment)
)
suppressPackageStartupMessages(
library(ComplexHeatmap)
)

Make the heatmap:

heatmap_se(se)

The defaults are very good.
The custom options are great.

Why SummarizedExperiment?

SummarizedExperiment is an enhanced data matrix

It stores multiple data matrices.
Some examples:
- Raw
- normalized
- batch-adjusted
- VST-normalized
- “counts” (or pseudocounts)
- “abundance” (or TPM, FPKM)
Columns have annotations in colData().
Rows have annotations in rowData().
Annotations are used in heatmap_se():
- display annotations beside the heatmap
- split the heatmap by groups
- center the data in meaningful ways
You can create a SummarizedExperiment from any of:
- numeric matrix
- ExpressionSet
- DGEList
- DESeqDataset

See the SummarizedExperiment vignette for more detail.

Heatmap Principles

The heatmap_se() is opinionated.

The default should be very close to ideal.
There are many custom options.

We build upon ComplexHeatmap::Heatmap().

It is amazing. And complex. We know this.
You can combine heatmaps:
- Add them + to display side-by-side.
- Use %v% to display top-to-bottom.

Data are assumed to be log2-transformed.

We recommend log2(1 + x), ymmv.
log2-transformed values are consistent with many
’Omics tools which report log2 fold change.

Data are not scaled. (Gasp)

We believe magnitude of change is important in Omics data.
It is helpful for critical review of the data,
which is the main purpose of a heatmap.
(A heatmap is not itself a decision-making tool,
however it very much helps assess the assumptions
and results obtained from other tools.)
Row-scaling adjusts the signal based upon the
variability on each row.
It is particularly effective when the range of
response is different for each row of data,
and where the range of response is more closely
associated with technical limitations than with
biological response.
In our experience, most ’Omics platforms are not dominated by this limitation.
Row-scaling has the nice benefit that it does not
require log2-transformed data.
It “naturally” transforms data into z-score units.
These units may or may not be biologically meaningful.
The z-score units are particularly useful when you
only want to show presence/absence, or high/low
relative to the range of signal on each row.
We aim to achieve many of the same goals without
scaling, because we find the actual magnitudes useful.

Data are row-centered.

The average signal of each row is subtracted from
that row, resulting “difference from average”.
We assume log2-transformed data, therefore values are
called “log2 differences”.
These log2 differences are commonly shown in volcano plots,
and in MA-plots.
The “log2 differences” can be converted to fold changes,
so we use fold change to label the color key.
We recommend using the heatmap to display data as
similar as possible to the underlying statistical tests,
consistent with other supporting figures.

Always use a divergent color scale with centered data.

(The same rule applies to z-scores, and correlations.)
The color scale uses Brewer colors 'RdBu' in reverse:
blue-white-red
and white is always at zero.
Red is Up because this is a HEAT map.
Blue is Down because it is cold.
If Blue were up, you should call it a COLD map.
A cold map can be useful, but it is not a heat map.

The color scale uses a fixed range.

Default range:
-8 to +8 fold.
It defines a consistent color-to-magnitude relationship.
Two-fold should look like a two-fold change.
This range represents biologically relevant changes for
most Omics data. (It can be adjusted.)
Extremely high values do not adjust the range,
otherwise it would compress the colors of everything else.
When the data have small changes, the heatmap will show
them as small changes.
The range may be adjusted when needed.

Data centering is flexible, and encouraged:

By default, centering uses all samples.
“Global centering”
Centering can be “versus controls”.
Centering can be within sub-groups, 'centerby groups'.
Examples of 'centerby groups':
- design factor(s),
- sample type,
- tissue type,
- cell line,
- processing batch,
- pairing factor,
Each 'centerby group' can have its own controls.
For example, you may center within each cell line,
then center each sample to its respective control group.

Annotations are “easy”:

top_colnames adds colData() annotations at the top.
rowData_colnames shows rowData() annotations on the left.
It will assign categorical colors, or scalar colors,
or divergent colors.
You can assign colors every to column individually.

Statistical hits are “easy”:

Provide sestats to add statistical hits to the left side.
Hits are colored by direction (Up/Down).
The heatmap will subset to show only these hits.

Rows and columns can be grouped:

row_split defines row groups
column_split defines column groups
use an integer number of groups, or annotation colnames().

Heatmap Walkthrough

Each step in the walkthrough adds some detail. * Sometimes we demonstrate other custom options as well,
just to give a flavor for the variety of styles available.

Test Data

We generate test data for this demonstration.

se <- make_se_test(nrow=1000, ngroups=4, nreps=8)

# optionally define factor levels to force the order of labels
rowData(se)$Class <- factor(
   sample(head(LETTERS, 5), size=nrow(se), replace=TRUE))

Basic Heatmap

Provide se (SummarizedExperiment data) and it will create the default heatmap. * top_colnames are auto-detected by default.

hm <- heatmap_se(se)
#> 'magick' package is suggested to install to give better rasterization.
#> 
#> Set `ht_opt$message = FALSE` to turn off this message.

# draw by printing hm, or call draw() to add useful options
draw(hm)

Custom Annotation Colors

Use sample_color_list to provide custom colors.
- Provide a list named by the annotation column.
- Or use colorjam::group2colors() to define categorical colors.
rowData_colnames added row annotations to the left.
left_annotation_name_rot=60 rotated the annotation label.

sample_color_list <- list(
   group=c(
      groupA="gold",
      groupB="darkorange2",
      groupC="firebrick3",
      groupD="darkorchid4"
   ),
   Class=colorjam::group2colors(
      unique(rowData(se)$Class)))

heatmap_se(se,
   rowData_colnames="Class",
   left_annotation_name_rot=60,
   sample_color_list=sample_color_list)
#> 'magick' package is suggested to install to give better rasterization.
#> 
#> Set `ht_opt$message = FALSE` to turn off this message.

Heatmap Legend Options

ComplexHeatmap::ht_opt() allows some stylistic options
to be default whenever you draw a heatmap.
Some common options:
- merge_legends=TRUE merges the color legends to one column.
- ROW_ANNO_PADDING, COLUMN_ANNO_PADDING defines a gap
  between heatmap and row/column annotations, respectively.
Also: show_left_annotation_name="top" places the left
annotation label at the top instead of the bottom.
Also: heatmap_se() will draw the heatmap by default,
useful when you don’t need to store the output.
(I usually end up needing the heatmap.)

ComplexHeatmap::ht_opt(merge_legends=TRUE)

heatmap_se(se,
   rowData_colnames="Class",
   show_left_annotation_name="top",
   left_annotation_name_rot=150,
   sample_color_list=sample_color_list)
#> 'magick' package is suggested to install to give better rasterization.
#> 
#> Set `ht_opt$message = FALSE` to turn off this message.

Row and Column Groups

Split Rows by Annotation

row_split="Class" will group rows by “Class”.

hm <- heatmap_se(se,
   rowData_colnames="Class",
   row_split="Class",
   sample_color_list=sample_color_list)
#> 'magick' package is suggested to install to give better rasterization.
#> 
#> Set `ht_opt$message = FALSE` to turn off this message.

draw(hm)

Split Rows and Columns

column_split="group" splits the columns.
row_split="Class" splits the rows.
You can provide multiple columns if needed.

hm <- heatmap_se(se,
   column_split="group",
   row_split="Class",
   rowData_colnames="Class",
   sample_color_list=sample_color_list)
#> 'magick' package is suggested to install to give better rasterization.
#> 
#> Set `ht_opt$message = FALSE` to turn off this message.

hm2 <- draw(hm)

Get the Row Group Order

Now seems like a good time to get the row order:
- jamba::heatmap_row_order(hm)
- It returns a list separated by row group.
It works best with the hm2 drawn heatmap, since
some clustering methods use a random element.

hro <- jamba::heatmap_row_order(hm2);
lengths(hro)
#>   A   B   C   D   E 
#> 196 210 199 215 180

See the first five entries per row group:

lapply(hro, head, 5)
#> $A
#>   row_0640   row_0405   row_0417   row_0730   row_0847 
#> "row_0640" "row_0405" "row_0417" "row_0730" "row_0847" 
#> 
#> $B
#>   row_0087   row_0075   row_0421   row_0624   row_0238 
#> "row_0087" "row_0075" "row_0421" "row_0624" "row_0238" 
#> 
#> $C
#>   row_0834   row_0593   row_0670   row_0823   row_0066 
#> "row_0834" "row_0593" "row_0670" "row_0823" "row_0066" 
#> 
#> $D
#>   row_0349   row_0755   row_0633   row_0784   row_0208 
#> "row_0349" "row_0755" "row_0633" "row_0784" "row_0208" 
#> 
#> $E
#>   row_0948   row_0115   row_0874   row_0165   row_0227 
#> "row_0948" "row_0115" "row_0874" "row_0165" "row_0227"

Add Grouped Column Labels

The example illustrates a two-factor design: 1. Genotype: wildtype (WT) and knockout (KO) 2. Treatment: Control and Dexamethasone (Dex)

Column split: column_split=c("Genotype", "Treatment")
controlSamples uses “WT” and “Control”
Several features are enabled:
- apply_hm_column_title=TRUE displays the heatmap title.
- top_colnames=FALSE hides the top annotation.
- heatmap_column_group_labels() adds grouped labels.
- hm_title_buffer adds whitespace for the grouped lines.
We add two design factors: Genotype, Treatment

colData(se)$Genotype <- rep(
   factor(c("WT", "KO"), levels=c("WT", "KO")),
   each=16);
colData(se)$Treatment <- rep(
   c("Control", "Dex"),
   each=8);

# center by groupA samples
use_controlSamples <- rownames(
   subset(colData(se),
      Genotype %in% "WT" &
      Treatment %in% "Control"))

hm9 <- heatmap_se(se,
   apply_hm_column_title=TRUE,
   hm_title_buffer=4,
   controlSamples=use_controlSamples,
   control_label="vs WT_Control",
   sestats=sestats,
   top_colnames=c("Genotype", "Treatment"),
   column_split=c("Genotype", "Treatment"),
   row_split=6,
   column_gap=grid::unit(c(1, 2, 1), "mm"),
   sample_color_list=sample_color_list)
hm9_drawn <- ComplexHeatmap::draw(hm9)

# now add fancy labels
heatmap_column_group_labels(
   hm_group_list=c("Treatment", "Genotype"),
   font_cex=1.3,
   se=se,
   hm_drawn=hm9_drawn)

Use y_offset_lines to adjust the y position of labels.

Change Column Group Order

The example below shows the same heatmap
grouped by: 1. Treatment 2. Genotype

hm10 <- heatmap_se(se,
   apply_hm_column_title=TRUE,
   hm_title_buffer=4,
   controlSamples=use_controlSamples,
   control_label="vs WT_Control",
   sestats=sestats,
   top_colnames=c("Treatment", "Genotype"),
   column_split=c("Treatment", "Genotype"),
   row_split=6,
   sample_color_list=sample_color_list)
hm10_drawn <- ComplexHeatmap::draw(hm10,
   merge_legends=TRUE)

# now add fancy labels
heatmap_column_group_labels(
   hm_group_list=c("Genotype", "Treatment"),
   font_cex=1.2,
   se=se,
   hm_drawn=hm10_drawn)

Data Centering

Default “data centering” subtracts the row mean value,
resulting in “difference from global mean”.
We refer to this approach as “global centering”.
Centering typically uses the mean value (the average),
however it can use median with useMedian=TRUE.
The median is effective in reducing the visual effects
of outlier points.
“Versus control”: Centering can be focused on a
control group, instead of global centering. This approach
is very effective at visualizing the data as it is seen
by statistical contrasts. In fact, it is encouraged.
“Within Groups”: Data can be centered within
sub-groups, called ‘center-by groups’. This approach
is quite useful and versatile.
1. Consider an experiment with two different cell lines,
  comparing treatment versus control.
  The cell lines have substantial differences from each other,
  however the treatment effects may be similar.
  We would “center-by Cell Line”, then “versus Control”.
2. Consider an experiment with cells derived from clinical
  patients, each patient sample is treated multiple ways.
  Again, “center-by Patient”, then either “global center” or
  “versus Control”.

Working examples are shown below.

Center Using a Control Group

controlSamples defines a subset of samples
to be the reference for data centering.
- Here we use samples in "groupA".
- control_label="vs GroupA" describes the control
column_title_rot=90 also rotates the column group
title 90 degrees.
We define a new attribute 'hm_title' as a
convenient heatmap title.
- column_title adds the title when we draw()
  the heatmap.

# center by groupA samples
use_controlSamples <- rownames(
      subset(colData(se), group %in% "groupA"))

# - control_label
hm3 <- heatmap_se(se,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   column_split="group",
   column_title_rot=90,
   row_split="Class",
   rowData_colnames="Class",
   cluster_row_slices=FALSE,
   sample_color_list=sample_color_list)

hm4 <- ComplexHeatmap::draw(hm3,
   column_title=attr(hm3, "hm_title"))

Center By Genotype

centerby_colnames="Genotype" centers within WT or KO
use_controlSamples_Treat uses ‘Control’ samples
in both WT and KO.
- It shows “Dex - Control” in each Genotype.
column_gap adds a custom gap between column groups.

use_controlSamples_Treat <- rownames(
      subset(colData(se), Treatment %in% "Control"))

hm11 <- heatmap_se(se,
   apply_hm_column_title=TRUE,
   hm_title_buffer=3,
   controlSamples=use_controlSamples_Treat,
   control_label="vs Control",
   sestats=sestats,
   top_colnames=FALSE,
   centerby_colnames="Genotype",
   column_split=c("Genotype", "Treatment"),
   row_split=6,
   column_gap=grid::unit(c(2, 4, 2), "mm"),
   sample_color_list=sample_color_list)
hm11_drawn <- ComplexHeatmap::draw(hm11,
   merge_legends=TRUE)

# now add fancy labels
heatmap_column_group_labels(
   hm_group_list=c("Treatment", "Genotype"),
   font_cex=1.2,
   se=se,
   hm_drawn=hm11_drawn)

Center by Batch

simulate_se_test()
- Simulates test data that contains a blocking effect.
  It also represents pairing factors (repeated measures), or batch effect.
- When simulating four groups (default),
  it simulates two-factor additive changes.
centerby_colnames="batch" > Important Note: This technique is entirely visual,
although it helps confirm potential structure within the data.
The math very closely mimics the adjustment performed in most
statistical tools, however we recommend using the proper
statistical adjustments, specifically block in limma, or
using an additional factor in DESeq2 or edgeR. It is generally
not recommended to perform statistical testing with
batch-adjusted data.

Without Batch-Centering

seb <- simulate_se_test(ngroups=4, nreps=8, multiplier=2)
#> Warning in rep(c("Ctl", "Dex"), each = nreps * 2): first element used of 'each'
#> argument
#> Warning in rep(c("WT", "KO"), each = nreps): first element used of 'each'
#> argument
#> Warning in matrix(ncol = ngroups, nrow = nrow(m), data = c(rep(0, nrow(m)), :
#> data length [1249] is not a sub-multiple or multiple of the number of rows
#> [250]
#> Warning in seq_len(nreps): first element used of 'length.out' argument
#> Warning in seq_len(nreps): first element used of 'length.out' argument
#> Warning in seq_len(nreps): first element used of 'length.out' argument
#> Warning in seq_len(nreps): first element used of 'length.out' argument
colData(seb)[, 2:3] <- colData(se)[, 3:4]
colnames(colData(seb))[2:3] <- colnames(colData(se))[3:4]

hm11 <- heatmap_se(
   seb,
   column_title="Global Centering",
   top_colnames=c("Genotype", "Treatment", "batch"),
   column_split=c("Genotype", "Treatment"),
   row_split=6,
   column_gap=grid::unit(c(2, 4, 2), "mm"),
   sample_color_list=sample_color_list)
hm11_drawn <- ComplexHeatmap::draw(hm11)

Center within batch

hm12 <- heatmap_se(
   seb,
   column_title="Within-Batch Centering",
   top_colnames=c("Genotype", "Treatment", "batch"),
   column_split=c("Genotype", "Treatment"),
   centerby_colnames="batch",
   row_split=6,
   column_gap=grid::unit(c(2, 4, 2), "mm"),
   sample_color_list=sample_color_list)
hm12_drawn <- ComplexHeatmap::draw(hm12)

Compare Two Heatmaps

ComplexHeatmap::Heatmap objects can be + added
together to display both heatmaps side-by-side.
**It requires both heatmaps to have identical rows.
- It does not require the heatmaps to have identical clustering, row split.
ht_gap uses grid::unit() to define a 3-cm
gap between heatmaps.

ComplexHeatmap::draw(
   hm11 + hm12,
   column_title="Comparing Two Data Centering Approaches",
   column_title_gp=grid::gpar(fontsize=20),
   ht_gap=grid::unit(3, "cm")
)
#> Warning: Heatmap/annotation names are duplicated: centered expression

Statistical Hits

Display Statistical Hits

sestats enables statistical hits on the left side.
It can be a list named by contrast, each with:
- a numeric vector named by gene rownames.
- values should be -1 down, or +1 up.
- Any non-zero value is considered a ‘hit’.
It can be an incidence matrix.
- rownames should match some heatmap rownames.
- colnames should be contrast names
- values are numeric, where -1 is down, +1 is up.
It can be SEStats from se_contrast_stats().
- This step is outside the scope, but is convenient!
- See Run Limma Statistics for an example.

We create a list of hit vectors:

# define a random set of hits
sestats_list <- list(
   contrast1=setNames(
      sample(c(1, -1), replace=TRUE, size=50),
      sample(rownames(se), size=50)
   ),
   contrast2=setNames(
      sample(c(1, -1), replace=TRUE, size=50),
      sample(rownames(se), size=50)
   )
)

Now create the heatmap with sestats:

hm6 <- heatmap_se(se,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   sestats=sestats_list,
   column_split="group",
   row_split="Class",
   rowData_colnames="Class",
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm6,
   column_title=attr(hm6, "hm_title"))

Run Limma Statistics

se_contrast_stats() automates limma for multiple contrasts.
- It applies lmFit(), fitContrasts(), topTable(),
  even voom() or DEqMS as appropriate.
- It analyzes each contrast and assay.
- It returns SEStats, used directly in the heatmap

# SEDesign
sedesign <- groups_to_sedesign(
   colData(se)[, "group", drop=FALSE])
# use only "versus groupA"
contrast_names(sedesign) <- jamba::vigrep("-groupA", contrast_names(sedesign))

# run stats
sestats <- se_contrast_stats(
   se=se,
   fold_cutoff=4,
   sedesign=sedesign,
   assay_name="counts")

Use SEStats with heatmap_se():

hm6s <- heatmap_se(se,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   sestats=sestats,
   column_split="group",
   row_split=6,
   rowData_colnames="Class",
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm6s,
   column_title=attr(hm6s, "hm_title"))

Drill Down by Row Cluster

row_subcluster="4" makes it easy to answer
“What is going on with Cluster 4?”

# for fun, "drill down" into cluster 5
hm6s_4 <- heatmap_se(se,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   sestats=sestats,
   column_split="group",
   row_split=6,
   row_subcluster=4,
   rowData_colnames="Class",
   sample_color_list=sample_color_list)
#> Warning: The heatmap has not been initialized. You might have different results
#> if you repeatedly execute this function, e.g. when row_km/column_km was
#> set. It is more suggested to do as `ht = draw(ht); row_order(ht)`.

ComplexHeatmap::draw(hm6s_4,
   column_title=attr(hm6s_4, "hm_title"))

Use a Hit Incidence Matrix

What is an incidence matrix, you wonder?
- We demonstrate the data format below.
- Rows are genes from rownames(se).
- Columns are contrasts, comparisons, or any useful label.
- Any non-zero value is considered a hit,
  and it colorized by direction.

# convert sestats to list, then incidence matrix
sestats_hitlist <- hit_array_to_list(sestats)
sestats_hitim <- venndir::list2im_value(sestats_hitlist[1:2])
print(head(sestats_hitim));
#>          groupB-groupA groupC-groupA
#> row_0022            -1            -1
#> row_0030            -1             0
#> row_0066             1             0
#> row_0075             1             0
#> row_0080            -1             0
#> row_0087             1             0

Now sestats_hitim defines statistical hits.

hm7 <- heatmap_se(se,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   sestats=sestats_hitim,
   column_split="group",
   rowData_colnames="Class",
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm7,
   column_title=attr(hm7, "hm_title"))

Correlation Heatmaps

Any standard heatmap_se() can be converted to a
correlation heatmap by adding:
correlation=TRUE

It uses the centered data that would have been displayed
in the heatmap, and instead calls cor() to create Pearson
correlation matrix.
- You can review the data used for correlation using:
  hm <- heatmap_se(se)
  then:
  hm@matrix
- Correlation uses use="pairwise.complete.obs" and ...
  ellipses can be used for custom method.
The heatmap defaults are changed:
- max_color=1 so the color range becomes -1 to +1
- row_split and row_label_colname refer to colData(se).
- legend_labels use legend_at directly, no fold conversion.
- cluster_columns controls rows and columns, for symmetry.
- The default legend title adds 'correlation of'.

hm8corr <- heatmap_se(seb,
   correlation=TRUE,
   apply_hm_column_title=TRUE,
   # controlSamples=use_controlSamples,
   cluster_columns=FALSE,
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm8corr)

Hint: This correlation heatmap shows what we expect
when there is a sample pairing or batch effect.
The key feature is “positive correlation along a diagonal.”

You can adjust for pairing factor using
centerby_colnames="batch":

hm8corr_batch <- heatmap_se(seb,
   correlation=TRUE,
   centerby_colnames="batch",
   apply_hm_column_title=TRUE,
   cluster_columns=FALSE,
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm8corr_batch)

Incidentally, the very small negative correlation along
the diagonal is also indicative of simple batch adjustment.
The common signal is removed across replicates in the same
batch, which by definition leaves only slight differences.
As a result, the residual is typically slightly negative.

Notice also that the overall group correlations have
increased substantially, due to the reduction in non-group
correlated signal. This increase is another indication that
the batch/pairing effect was appropriately modeled.

Custom Visual Options

Add Callout Labels for Some Rows

mark_rows adds labels to the right
- Great for labeling genes of interest, especially
  when there are too many rows in the heatmap.
- We add 5 labels from ‘A’, and 3 labels from ‘D’.
In RStudio, you may want to silence some warnings:
ComplexHeatmap::ht_opt(message=FALSE)

# add "callout" labels for a subset of rows
hro <- jamba::heatmap_row_order(hm4);
mark_rows <- c(
   sample(hro[["A"]], size=5),
   sample(hro[["D"]], size=3));

hm5 <- heatmap_se(se,
   mark_rows=mark_rows,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   column_split="group",
   column_title_rot=90,
   row_split="Class",
   rowData_colnames="Class",
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm5,
   column_title=attr(hm5, "hm_title"))

Customize Column Labels

Many labels options can be customized: size, color, fontface
- This example shows how to change the color
  using column_names_gp, which uses grid::gpar().

# customize column label fonts using column_names_gp
column_bold <- ifelse(
   SummarizedExperiment::colData(se)$group %in% "groupA",
   2, 1);
column_colors <- sample_color_list$group[
   as.character(SummarizedExperiment::colData(se)$group)
]

hm8 <- heatmap_se(se,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   column_names_gp=grid::gpar(
      col=column_colors,
      font=column_bold
   ),
   column_split="group",
   row_split="Class",
   rowData_colnames="Class",
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm8,
   column_title=attr(hm8, "hm_title"))

Create a Correlation Heatmap

Just add correlation=TRUE.
- Any heatmap becomes a correlation heatmap,
  using the same centered data as displayed
  in the normal heatmap.
- Centered data are used with cor(), which is
  the recommended way to calculate correlation values.
- It answers “What correlates using difference-from-average?”
cluster_columns=TRUE enables column clustering.
Remove column_split to allow unsupervised clustering.

# correlation=TRUE, any heatmap becomes a sample correlation heatmap
hm8corr <- heatmap_se(se,
   correlation=TRUE,
   apply_hm_column_title=TRUE,
   controlSamples=use_controlSamples,
   control_label="vs groupA",
   column_names_gp=grid::gpar(
      col=column_colors,
      font=column_bold),
   cluster_columns=TRUE,
   sample_color_list=sample_color_list)

ComplexHeatmap::draw(hm8corr)