column_order is ignored when column_split is used

[dcarbajo]

I have a heat map like this, where I apply column_order to obtain the order of columns that I want (with NoT group before LTR group):

Now I just want to apply a column gap between the NoT and the LTR columns. For that I use column_split with a variable that maps my matrix column names to their group, in the same order.

This worked for me before, cause my groups were A and B, so the A-B order was kept. However, now I want to plot this groups as B-A, so when I apply column_split, the group order is reversed, ignoring my column_order.

How can I overcome this? I tried changing the order of my split variable, but to no avail. Thanks!

[dcarbajo]

I posted a full MWE in here, pasted below as well:

Check the following MWE with the iris data.

1- Create my data matrix (small subset of iris for just setosa and virginica Species), and a meta information data frame with just sample ID and Species (my grouping variable):

data(iris)
my_setosa=subset(iris, Species=="setosa")
my_virginica=subset(iris, Species=="virginica")
set.seed(123)
rows_used1 <- sort(sample(1:nrow(my_setosa), 5, replace = F))
rows_used2 <- sort(sample(1:nrow(my_virginica), 5, replace = F))
my_iris=rbind(my_setosa[rows_used1,], my_virginica[rows_used2,])
#
heat_mat=t(as.matrix(my_iris[,-ncol(my_iris)]))
meta_df=data.frame(ID=paste0("id",rownames(my_iris)), Species=my_iris[,ncol(my_iris)])
meta_df$Species=droplevels(meta_df$Species)
colnames(heat_mat)=meta_df$ID

These look like this:

> heat_mat
             id3 id14 id15 id31 id42 id114 id125 id137 id143 id150
Sepal.Length 4.7  4.3  5.8  4.8  4.5   5.7   6.7   6.3   5.8   5.9
Sepal.Width  3.2  3.0  4.0  3.1  2.3   2.5   3.3   3.4   2.7   3.0
Petal.Length 1.3  1.1  1.2  1.6  1.3   5.0   5.7   5.6   5.1   5.1
Petal.Width  0.2  0.1  0.2  0.2  0.3   2.0   2.1   2.4   1.9   1.8
> meta_df
      ID   Species
1    id3    setosa
2   id14    setosa
3   id15    setosa
4   id31    setosa
5   id42    setosa
6  id114 virginica
7  id125 virginica
8  id137 virginica
9  id143 virginica
10 id150 virginica

2- Define heatmap and Species grouping colors, column annotation (Species groups), column order and column split (gap between Species groups):

palette=grDevices::colorRampPalette(c("green","black","red"))(11)
col_vec=c("red","blue")
col_vec=stats::setNames(col_vec, levels(meta_df$Species))
column_ha <- ComplexHeatmap::HeatmapAnnotation(
               Species = meta_df$Species,
               col = list(Species = col_vec),
               show_legend = TRUE)
column_order=meta_df[order(meta_df$Species, decreasing=T),]$ID
column_split <- meta_df$Species

Note here (and this is the problem), that I want the virginica group on the left, and the setosa group on the right, so my column order is the following:

> column_order
 [1] "id114" "id125" "id137" "id143" "id150" "id3"   "id14"  "id15"  "id31"
[10] "id42"

3- Make the heatmap without column_split:

complex_heat <- ComplexHeatmap::Heatmap(heat_mat, cluster_rows = FALSE, cluster_columns = FALSE,
                                        row_dend_width = grid::unit(2, "inch"),
                                        rect_gp = grid::gpar(col = "white", lwd = 2),
                                        col = palette,
                                        top_annotation = column_ha,
                                        column_order = column_order,
                                        #column_split = column_split,
                                        column_gap = grid::unit(0.1, "inch"),
                                        border = TRUE)
grDevices::png(filename="heatmap.png", height=400, width=600)
ComplexHeatmap::draw(complex_heat)
grDevices::dev.off()

This is all good, the heatmap produced below has my sample IDs ordered correctly with virginica on the left (in blue as specified by col_vector), and setosa on the right (in red as specified by col_vector):

4- Make the heatmap with column_split; if I just uncomment the column_split line that specifies to split the column by the Species variable, I woould expect the exact same heatmap, with just a gap between the two Species groups. However, the column_order is ignored, and setosa samples appear on the left...

complex_heat <- ComplexHeatmap::Heatmap(heat_mat, cluster_rows = FALSE, cluster_columns = FALSE,
                                        row_dend_width = grid::unit(2, "inch"),
                                        rect_gp = grid::gpar(col = "white", lwd = 2),
                                        col = palette,
                                        top_annotation = column_ha,
                                        column_order = column_order,
                                        column_split = column_split,
                                        column_gap = grid::unit(0.1, "inch"),
                                        border = TRUE)

[dcarbajo]

Found the answer, column_split should be a factor with the same order as meta_df$Species, but the levels specified in the order we want them plotted...

The solution is to define column_split like this:

column_split=factor(as.character(meta_df$Species), levels=c('virginica','setosa'))

[hookoop]

Hey,
I'm writing this for anyone who stumbles upon this thread because I could not get this solution to work.. I could get the columns to split how I want to, but the samples get suffled so that they do not correspond to the assigned groups. I have specified the levels of my factor, and factor(meta_df$treat, levels= unique(meta_df$treat)) has always worked for me before.

The only solution I could find was column_split = factor(c(2,2,2,1)) , as my object had levels "PRE" (pre treatment) and "POST" (post treatment) corresponding to 1 and 2 respectively. If I try to order the column_split by any variation of c("PRE", "POST","POST","POST") I will not get my desired outcome. This is still confusing because on the heatmap the order shows up as 1, 2,2,2 (like I wanted) yet I has to specify it to be 2,2,2,1 in the code.

[dcarbajo]

See if any of these 2 questions in stackoverflow help you out:

https://stackoverflow.com/questions/78099435/complexheat-column-order-is-ignored-when-column-split-is-used

https://stackoverflow.com/questions/78184614/r-complexheatmap-cannot-reproduce-exact-row-orders-when-apply-row-clusters-to-ne

[vetmohit89]

I tried above suggested method to re-order the columns but no help.

Heatmap(Actinobacteria_counts_scaled_matrix_clean_sorted, cluster_columns = dend2, column_split = factor(as.character(meta_df$Species), levels=c('WT','db/db', 'db+ACE2', 'db+IW')),
+         column_order=column_order,
+         cluster_row_slices = TRUE,
+         cluster_column_slices = TRUE,
+          top_annotation = HeatmapAnnotation(foo = fa, col = list(foo = fa_col)))

fa = rep(c("WT", "DB", "DB_ACE", "DB_IW"), times = c(9, 7, 4, 5))
fa_col <- c("WT" = 2, "DB" = 3, "DB_ACE" = 4, "DB_IW" = 5)

dend2 = cluster_within_group(Actinobacteria_counts_scaled_matrix_clean, fa)

Error: When cluster_columns is a dendrogram, column_split can only be a single number.