Heatmap.2 giving specific color to a value and start gradient after

Question

[Seigfried]

I am trying to plot a Heatmap. My data is as follows.

GO  s_3d    s_6d    s_17d
peptide metabolic process   4.71E-29    2.84828E-20 2.70719E-24
organonitrogen compound biosynthetic process    3.38158E-22 3.26674E-15 6.3903E-22
positive regulation of gene expression      0.057421474 
negative regulation of RNA metabolic process        0.05933718  
G-protein coupled receptor signaling pathway        0.061777665 
positive regulation of macromolecule biosynthetic process       0.061906369 
negative regulation of nitrogen compound metabolic process      0.065645467

Basically the data has 4 columns and a LOT of NA values. I am using the function Heatmap.2 which usually has an option called na.color which can help me color the NA values to a distinct color

Since I have too many NA values, the heatmap function gives me an error

Error in hclustfun(distr) : NA/NaN/Inf in foreign function call (arg 11)

Therefore I tried to replace all NAs with zero and then give that a separate color. However the heatmap gives me a gradient even though I give the color separate. What I want is all 0's to be White and Everything which is not 0 to have a gradient from yellow to red. I do not want a gradient between white to yellow. How do I do this?

This is my code

rm(list = ls())
cat("\014") 

if (!require("RColorBrewer")) {
install.packages("RColorBrewer", dependencies = TRUE)
library(RColorBrewer)
}

file <- "C:/Users/niran/Downloads/go_cluster.txt"   #input
data <- read.table(file,header=T, sep = "\t",stringsAsFactors = F)
row.names(data) <- data[,1]
data <- data[-1]

# Since data is spread too far apart we can do a log transform to reduce the gaps

data[is.na(data)] <- 0 # Find possible solution for this
data <- -log10(data + 0.000000000001)

# creates a own color palette from yellow to red
color.palette  <- colorRampPalette(c("#F8F8F8","yellow", "orange", "red"))(n=600)

col_breaks = c(seq(0,0.1,length=1),  # for white
           seq(0.1,2,length=100),
           seq(2.01,4,length=100),
           seq(4.01,6,length=100),
           seq(6.01,8,length=100),
           seq(8.01,10,length=100),
           seq(10.01,12,length=100))

Heatmap <- heatmap.2(as.matrix(data),
      main = "Heatmap of p-values for GO",
      density.info = "none",
      trace = "none",
      margins = c(5,28),
      key.xlab = "log10 Values",
      cexRow = 1,
      cexCol = 1.5,
      keysize=0.75,
      col = color.palette,  # use on color palette defined earlier
      breaks=col_breaks,    # enable color transition at specified limits
      dendrogram ="both")

dev.off()

Answer 1

[e.rempel]

Hi Seigfried,

you give the separate colors, but ColorRamPalette still interpolates them to create a smooth gradient from white to yellow to orange to red. If I understood you correctly, you would like first create a smooth gradient from yellow to orange to red and then prepend "white". Thus your code should contain something like:

color.palette  <- c("#F8F8F8", colorRampPalette(c("yellow", "orange", "red"))(n=599))

[Seigfried]

Yes this is it! Didnt think of that. Thank you!

Answer 2

[Samuel Brady]

Are you certain that having too many NA values is really causing this error?

According to this post, you may have an infinite value:

https://stackoverflow.com/questions/23738900/error-while-creating-heatmaps-na-nan-inf-in-foreign-function-call-arg-11

Answer 3

[Seigfried]

Hi Samuel

any( sapply(data, is.infinite) )
[1] FALSE
any( sapply(data, is.na) )
[1] TRUE
any( sapply(data, is.nan) )
[1] FALSE

There is no infinite value or NAN in the matrix. I thought that the hclust cannot plot the distance between samples because of too many NAs

[h.mon]

Do not answer here to an answer to your question, use the ADD COMMENT button at the appropriate answer.