Difficulty positioning heatmap.2 components - r

Question

I have been really struggling to position the components of my heatmap.2 output.
I found this old answer explaining how the element positioning worked from #IanSudbery which seemed really clear and I thought it had given me the understanding I need, but I'm still not grasping something.
I understand that the elements are all essentially put in a lattice of windows but they aren't behaving in a way I understand.
Here is my code and the current output (at the very bottom is the bit of interest which orders the figure elements):
for(i in 1:length(ConditionsAbbr)) {
# creates its own colour palette
my_palette <- colorRampPalette(c("snow", "yellow", "darkorange", "red"))(n = 399)
# (optional) defines the colour breaks manually for a "skewed" colour transition
col_breaks = c(seq(0,0.09,length=100), #white 'snow'
seq(0.1,0.19,length=100), # for yellow
seq(0.2,0.29,length=100), # for orange 'darkorange'
seq(0.3,1,length=100)) # for red
# creates a 5 x 5 inch image
png(paste(SourceDir, "Heatmap_", ConditionsAbbr[i], "XYZ.png"), # create PNG for the heat map
width = 5*600, # 5 x 600 pixels
height = 5*600,
res = 300, # 300 pixels per inch
pointsize = 8) # smaller font size
heatmap.2(ConditionsMtx[[ConditionsAbbr[i]]],
cellnote = ConditionsMtx[[ConditionsAbbr[i]]], # same data set for cell labels
main = paste(ConditionsAbbr[i], "XYZ"), # heat map title
notecol="black", # change font color of cell labels to black
density.info="none", # turns off density plot inside color legend
trace="none", # turns off trace lines inside the heat map
margins =c(12,9), # widens margins around plot
col=my_palette, # use on color palette defined earlier
breaks=col_breaks, # enable color transition at specified limits
dendrogram="none", # No dendogram
srtCol = 0 , #correct angle of label numbers
asp = 1 , #this overrides layout methinks and for some reason makes it square
adjCol = c(NA, -35) ,
adjRow = c(53, NA) ,
keysize = 1.2 ,
Colv = FALSE , #turn off column clustering
Rowv = FALSE , # turn off row clustering
key.xlab = paste("Correlation") ,
lmat = rbind( c(0, 3), c(2,1), c(0,4) ),
lhei = c(0.9, 4, 0.5) )
dev.off() # close the PNG device
}
This gives:
As you can see, the key is right of the matrix, there are huge amounts of white space between the matrix, the title above and key below, and it's not even as if the title and matrix are centred in the PNG?
I think to myself "well I'll just create a 3x3 that is easy to understand and edit" e.g.
| |
| | (3)
| |
--------------------------
| (1) |
(2) | Matrix |
| |
--------------------------
| (4) |
| Key |
| |
And then I can get rid of the white space so it's more like this.
| |(3)
------------------
| (1) |
(2)| Matrix |
| |
------------------
|(4) Key |
I do this using:
lmat = rbind( c(0, 0, 3), c(2, 1, 0), c(0, 4, 0) ),
lhei = c(0.9, 4, 0.5) ,
lwid = c(1, 4, 1))
This is what it looks like:
As great as it is to see my matrix in the centre, my key is still aligned to the right of my matrix and my title is taking the Silk Road East? Not to mention all the excess white space?
How do I get these to align and to all move together so the figure components fit snugly together?
EDIT: reducing my margins helped to reduce the whitespace but it's still excessive.

Answer 1

Here are the final changes I made to get my results, however, I would recommend using the advice of Maurits Evers if you aren't too invested in heatmap.2. Don't overlook the changes I made to the image dimensions.
# creates my own colour palette
my_palette <- colorRampPalette(c("snow", "yellow", "darkorange", "red"))(n = 399)
# (optional) defines the colour breaks manually for a "skewed" colour transition
col_breaks = c(seq(0,0.09,length=100), #white 'snow'
seq(0.1,0.19,length=100), # for yellow
seq(0.2,0.29,length=100), # for orange 'darkorange'
seq(0.3,1,length=100)) # for red
# creates an image
png(paste(SourceDir, "Heatmap_XYZ.png" )
# create PNG for the heat map
width = 5*580, # 5 x 580 pixels
height = 5*420, # 5 x 420 pixels
res = 300, # 300 pixels per inch
pointsize =11) # smaller font size
heatmap.2(ConditionsMtx[[ConditionsAbbr[i]]],
cellnote = ConditionsMtx[[ConditionsAbbr[i]]], # same data set for cell labels
main = "XYZ", # heat map title
notecol="black", # change font color of cell labels to black
density.info="none", # turns off density plot inside color legend
trace="none", # turns off trace lines inside the heat map
margins=c(0,0), # widens margins around plot
col=my_palette, # use on color palette defined earlier
breaks=col_breaks, # enable color transition at specified limits
dendrogram="none", # only draw a row dendrogram
srtCol = 0 , #correct angle of label numbers
asp = 1 , #this overrides layout methinks and for some reason makes it square
adjCol = c(NA, -38.3) , #shift column labels
adjRow = c(77.5, NA) , #shift row labels
keysize = 2 , #alter key size
Colv = FALSE , #turn off column clustering
Rowv = FALSE , # turn off row clustering
key.xlab = paste("Correlation") , #add label to key
cexRow = (1.8) , # alter row label font size
cexCol = (1.8) , # alter column label font size
notecex = (1.5) , # Alter cell font size
lmat = rbind( c(0, 3, 0), c(2, 1, 0), c(0, 4, 0) ) ,
lhei = c(0.43, 2.6, 0.6) , # Alter dimensions of display array cell heighs
lwid = c(0.6, 4, 0.6) , # Alter dimensions of display array cell widths
key.par=list(mar=c(4.5,0, 1.8,0) ) ) #tweak specific key paramters
dev.off()
Here is the output, which I will continue to refine until all spacing and font sizes suit my aesthetic preference. I would tell you exactly what I've done but I'm not 100% sure, frankly it all feels like it's held together with old gum and bailer twine, but don't kick a gift horse in the code, as they say.

Answer 2

I don't know if you're open to non-heatmap.2-based solutions. In my opinion ggplot offers greater flexibility and with a bit of tweaking you can reproduce a heatmap similar to the one you're showing quite comfortably while maximising plotting "real-estate" and avoiding excessive whitespace.
I'm happy to remove this post if you're only looking for heatmap.2 solutions.
That aside, a ggplot2 solution may look like this:
First off, let's generate some sample data
set.seed(2018)
df <- as_tibble(matrix(runif(7*10), ncol = 10), .name_repair = ~seq(1:10))
Prior to plotting we need to reshape df from wide to long
library(tidyverse)
df <- df %>%
rowid_to_column("row") %>%
gather(col, Correlation, -row) %>%
mutate(col = as.integer(col))
Then to plot
ggplot(df, aes(row, col, fill = Correlation)) +
geom_tile() +
scale_fill_gradientn(colours = my_palette) + # Use your custom colour palette
theme_void() + # Minimal theme
labs(title = "Main title") +
geom_text(aes(label = sprintf("%2.1f", Correlation)), size = 2) +
theme(
plot.title = element_text(hjust = 1), # Right-aligned text
legend.position="bottom") + # Legend at the bottom
guides(fill = guide_colourbar(
title.position = "bottom", # Legend title below bar
barwidth = 25, # Extend bar length
title.hjust = 0.5))
An example with multiple heatmaps in a grid layout via facet_wrap
First off, let's generate more complex data.
set.seed(2018)
df <- replicate(
4,
as_tibble(matrix(runif(7*10), ncol = 10), .name_repair = ~seq(1:10)), simplify = F) %>%
setNames(., paste("data", 1:4, sep = "")) %>%
map(~ .x %>% rowid_to_column("row") %>%
gather(col, Correlation, -row) %>%
mutate(col = as.integer(col))) %>%
bind_rows(.id = "data")
Then the plotting is identical to what we did before plus an additional facet_wrap(~data, ncol = 2) statement
ggplot(df, aes(row, col, fill = Correlation)) +
geom_tile() +
scale_fill_gradientn(colours = my_palette) + # Use your custom colour palette
theme_void() + # Minimal theme
labs(title = "Main title") +
geom_text(aes(label = sprintf("%2.1f", Correlation)), size = 2) +
facet_wrap(~ data, ncol = 2) +
theme(
plot.title = element_text(hjust = 1), # Right-aligned text
legend.position="bottom") + # Legend at the bottom
guides(fill = guide_colourbar(
title.position = "bottom", # Legend title below bar
barwidth = 25, # Extend bar length
title.hjust = 0.5))
One final update
I thought it'd be fun/interesting to see how far we can get towards a complex heatmap similar to the one you link to from the paper.
The sample data is included at the end, as this takes up a bit of space.
We first construct three different ggplot2 plot objects that show the main heatmap (gg3), an additional smaller heatmap with missing values (gg2), and a bar denoting group labels for every row (gg1).
gg3 <- ggplot(df.cor, aes(col, row, fill = Correlation)) +
geom_tile() +
scale_fill_distiller(palette = "RdYlBu") +
theme_void() +
labs(title = "Main title") +
geom_text(aes(label = sprintf("%2.1f", Correlation)), size = 2) +
scale_y_discrete(position = "right") +
theme(
plot.title = element_text(hjust = 1),
legend.position="bottom",
axis.text.y = element_text(color = "black", size = 10)) +
guides(fill = guide_colourbar(
title.position = "bottom",
barwidth = 10,
title.hjust = 0.5))
gg2 <- ggplot(df.flag, aes(col, row, fill = Correlation)) +
geom_tile(colour = "grey") +
scale_fill_distiller(palette = "RdYlBu", guide = F, na.value = "white") +
theme_void() +
scale_x_discrete(position = "top") +
theme(
axis.text.x = element_text(color = "black", size = 10, angle = 90, hjust = 1, vjust = 0.5))
gg1 <- ggplot(df.bar, aes(1, row, fill = grp)) +
geom_tile() +
scale_fill_manual(values = c("grp1" = "orange", "grp2" = "green")) +
theme_void() +
theme(legend.position = "left")
We can now use egg::ggarrange to position all three plots such that the y axis ranges are aligned.
library(egg)
ggarrange(gg1, gg2, gg3, ncol = 3, widths = c(0.1, 1, 3))
Sample data
library(tidyverse)
set.seed(2018)
nrow <- 7
ncol <- 20
df.cor <- matrix(runif(nrow * ncol, min = -1, max = 1), nrow = nrow) %>%
as_tibble(.name_repair = ~seq(1:ncol)) %>%
rowid_to_column("row") %>%
gather(col, Correlation, -row) %>%
mutate(
row = factor(
paste("row", row, sep = ""),
levels = paste("row", 1:nrow, sep = "")),
col = factor(
paste("col", col, sep = ""),
levels = paste("col", 1:ncol, sep = "")))
nrow <- 7
ncol <- 10
df.flag <- matrix(runif(nrow * ncol, min = -1, max = 1), nrow = nrow) %>%
as_tibble(.name_repair = ~seq(1:ncol)) %>%
rowid_to_column("row") %>%
gather(col, Correlation, -row) %>%
mutate(
row = factor(
paste("row", row, sep = ""),
levels = paste("row", 1:nrow, sep = "")),
col = factor(
paste("col", col, sep = ""),
levels = paste("col", 1:ncol, sep = ""))) %>%
mutate(Correlation = ifelse(abs(Correlation) < 0.5, NA, Correlation))
df.bar <- data.frame(
row = 1:nrow,
grp = paste("grp", c(rep(1, nrow - 3), rep(2, 3)), sep = "")) %>%
mutate(
row = factor(
paste("row", row, sep = ""),
levels = paste("row", 1:nrow, sep = "")))