Skip to content

@dsparks /Heatmap.R
Last active

Embed URL


Subversion checkout URL

You can clone with
Download ZIP
ggplot2 heatmap with "spectral" palette
doInstall <- TRUE # Change to FALSE if you don't want packages installed.
toInstall <- c("ggplot2", "reshape2", "RColorBrewer")
if(doInstall){install.packages(toInstall, repos = "")}
lapply(toInstall, library, character.only = TRUE)
# Generate a random matrix
# This can be any type of numeric matrix,
# though we often see heatmaps of square correlation matrices.
nRow <- 9
nCol <- 16
myData <- matrix(rnorm(nRow * nCol), ncol = nCol)
rownames(myData) <- letters[1:nRow]
colnames(myData) <- LETTERS[1:nCol]
# Replace with numbers that actually have a relationship:
for(ii in 2:ncol(myData)){ myData[, ii] <- myData[, ii-1] + rnorm(nrow(myData)) }
for(ii in 2:nrow(myData)){ myData[ii, ] <- myData[ii-1, ] + rnorm(ncol(myData)) }
# For melt() to work seamlessly, myData has to be a matrix.
longData <- melt(myData)
head(longData, 20)
# Optionally, reorder both the row and column variables in any order
# Here, they are sorted by mean value
longData$Var1 <- factor(longData$Var1, names(sort(with(longData, by(value, Var1, mean)))))
longData$Var2 <- factor(longData$Var2, names(sort(with(longData, by(value, Var2, mean)))))
# Define palette
myPalette <- colorRampPalette(rev(brewer.pal(11, "Spectral")), space="Lab")
zp1 <- ggplot(longData,
aes(x = Var2, y = Var1, fill = value))
zp1 <- zp1 + geom_tile()
zp1 <- zp1 + scale_fill_gradientn(colours = myPalette(100))
zp1 <- zp1 + scale_x_discrete(expand = c(0, 0))
zp1 <- zp1 + scale_y_discrete(expand = c(0, 0))
zp1 <- zp1 + coord_equal()
zp1 <- zp1 + theme_bw()

It should be:
longData$X1 <- factor(longData$Var1, names(sort(with(longData, by(value, Var1, mean)))))
longData$X2 <- factor(longData$Var2, names(sort(with(longData, by(value, Var2, mean)))))

Instead of:
longData$X1 <- factor(longData$X1, names(sort(with(longData, by(value, X1, mean)))))
longData$X2 <- factor(longData$X2, names(sort(with(longData, by(value, X2, mean)))))


Fixed, thanks!


This is cool @dsparks. I'm working on Plotly and we're working on translating ggplot2 plots into interactive, web-based plots. We did it with this one here:

You add py$ggplotly() to the end of the call to get it. See: for more. Very cool plot!


Nowadays, melt() generates "Var1" and "Var2" column names, instead of "X1" and "X2"


Alternatively, use rje::cubeHelix(100) in place of myPalette(100).


simple execution gives the following error:
Error in by.default(value, Var1, mean) : object 'Var1' not found
Am I doing anything wrong ?

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment
Something went wrong with that request. Please try again.