Question

color scale in cnetplot (enrichplot) is not centered in 0, skewed towards positive values

0

Entering edit mode

Emilia ▴ 40

@emiliabaffo

Last seen 3.5 years ago

Argentina, Rosario (Universidad Naciona…

Hello! I performed a pathway enrichment analysis using clusterProfiler and I wanted to plot a gene network with it using the cnetplot function from the enrichplot package. However, I can't get the 0 to be in the middle of the color scale and this causes some confusion in the visual interpretation of the plot. I've seen that this topic has been already adressed on github (here and here, for example) but I honestly still don't understand what the actual solution is. Could someone post an actual example of how the code for that would look like? This is what I've done so far.

p1 <- cnetplot(pathways, 
         showCategory = 10, 
         foldChange = geneList, 
         node_label="category", 
         categorySize="pvalue")
p1 + scale_color_gradientn(colours = c("green", "red"))

And this is how the plot looks, you can see that the color scale is unevenly distributed (I'll fix the overlapping node labels later, I'm more concerned about the color issue now) (Also, if anyone knows how to get rid of that "as.numeric(as.character(color))" that appears on top of the color scale, that would be a bonus track) Thanks in advance! cnet plot

enrichplot clusterProfiler • 7.6k views

ADD COMMENT • link 3.9 years ago • updated 3.8 years ago Emilia ▴ 40

0

Entering edit mode

What exactly do you mean by " the color scale is unevenly distributed"? I am asking because you (apparently) use fold-changes as input for color coding, and these values (in principle) vary between 0 and infinity... What does range(geneList) give? Also, "to get rid of that "as.numeric(as.character(color))"" you can use the argument name in the function scale_color_gradientn(). E.g.: p1 + scale_color_gradientn(name = "fold change", colours = c("green", "red"))

ADD REPLY • link 3.9 years ago Guido Hooiveld ★ 4.1k

0

Entering edit mode

Thanks for the advice on that! As to the fold changes, the maximum value is around +12 and the minimum value about -5. What I mean is that the color in the middle of red and green isn't 0, so this makes the visual interpretation of the graph a little misleading. You can see in the image that the intermediate color between red and green corresponds to a fold change of +4, when the most intuitive thing would be that this value were 0. I am aware that the reason for this must be that I have a greater range towards the positive values rather than the negatives, but I would still like the center to be at 0. I hope I'm making myself clear.

ADD REPLY • link 3.9 years ago • updated 3.8 years ago Emilia ▴ 40

1

Entering edit mode

So if I understand correctly, you are not happy with the default color gradient cnetplot uses (i.e. blue-white-red), but rather would like to use green-red. Please note that graphs made using green-red cannot properly be interpreted by people that are color-blind, but blue-white-red graphs (apparently) can (and this is why it is the default color used by cnetplot). See e.g. here.

Still, if you would like to go for a green-red color coding, why not include 'white' and use the function scale_colour_gradient2()? To do this:

p1 + scale_colour_gradient2(name = "fold change", low = "green", mid = "white", high = "red").

The benefit of doing so is that the color white then represents 0, which clearly stands out.

If you still would like to go with using the function scale_color_gradientn() like you did, then you can first extract the fold changes values of the genes to be plotted from the object p1 (i.e. the output of cnetplot) by p1$data$color (the slot data contains all relevant data used to make the plot, and has the advantage that it is limited to only the genes that are in the network, and not all genes present in geneList). Next determine the min and max values of the fold changes, and round it (better use for the latter floor and ceiling and not round):

min.value <- floor( min(p1$data$color, na.rm = TRUE) )
max.value <- ceiling( max(p1$data$color, na.rm = TRUE) )

Then use the arguments limits and breaks to make the plot. Note that in breaks zero (0) is added. You can of course also set the limits manually.

p1 + scale_color_gradientn(name = "fold change",
    colours = c("green", "red"),
    limits= c(min.value, max.value), 
    breaks=c(min.value , 0, max.value) )

Added: Note that you can also use limits and breaks with the function scale_colour_gradient2():

# manually defined values
min.value =-5
max.value = 5


p1 + scale_colour_gradient2(name = "fold change",
    low = "green", mid = "white", high = "red",
    limits= c(min.value, max.value),
    breaks=c(min.value , 0, max.value) )

ADD REPLY • link 3.9 years ago Guido Hooiveld ★ 4.1k

0

Entering edit mode

In fact when I don't indicate any particular color scale, the default is white-red and not blue-red, that's why I wished to change it since white-red is not very nice to interpret. I chose green-red for no particular reason but of course I would have no issue in using a colorblind friendly scale. I tried to do it this way as you suggested

p1 + scale_colour_gradient2(name = "fold change", low = "blue", mid = "white", high = "red")

But the colors remained the same as the default (white-red), it was as if I hadn't changed anything. So I ran this other code that you also suggested

min.value <- floor( min(p1$data$color, na.rm = TRUE) )
max.value <- ceiling( max(p1$data$color, na.rm = TRUE) )
p1 + scale_color_gradientn(name = "fold change",
                       colours = c("blue", "white", "red"),
                       limits= c(min.value, max.value), 
                       breaks=c(min.value , 0, max.value))

and I was able to set the color scale but I can't get the 0 to appear as white, which is what I wanted in order to make it easier to interpret visually. color bar plot

I suppose that it's because I have a wider range of values towards the positive values than towards the negative. But despite this fact, I'd still like the color scale to be even for both sides, so that white is no change, red is upregulated and blue is downregulated.

ADD REPLY • link 3.9 years ago Emilia ▴ 40

1

Entering edit mode

OK, what you then will need is the rescale function from the library scales (to scale the values). From the help pages ?scale_color_gradientn

for values:

values: if colours should not be evenly positioned along the gradient this vector gives the position (between 0 and 1) for each colour in the colours vector. See rescale() for a convenience function to map an arbitrary range to between 0 and 1.

library(scales)

# blue-white-red colour scale
p1 + scale_color_gradientn(name = "fold change",
        colours = c("blue","white","red"), 
        values = rescale(c(min.value, 0, max.value)),
        limits=c(min.value, max.value),
        breaks=c(min.value , 0, max.value) )

#red-green colour scale
p1 + scale_color_gradientn(name = "fold change",
    colours = c("green", "red"),
    values = rescale(c(min.value, 0, max.value)),
    limits= c(min.value, max.value), 
    breaks=c(min.value , 0, max.value) )