Stacked histograms like in flow cytometry

Question

I'm trying to use ggplot or base R to produce something like the following:

I know how to do histograms with ggplot2, and can easily separate them using facet_grid or facet_wrap. But I'd like to "stagger" them vertically, such that they have some overlap, as shown below. Sorry, I'm not allowed to post my own image, and it's quite difficult to find a simpler picture of what I want. If I could, I would only post the top-left panel.

I understand that this is not a particularly good way to display data -- but that decision does not rest with me.

A sample dataset would be as follows:

my.data <- as.data.frame(rbind( cbind( rnorm(1e3), 1) , cbind( rnorm(1e3)+2, 2), cbind( rnorm(1e3)+3, 3), cbind( rnorm(1e3)+4, 4)))

And I can plot it with geom_histogram as follows:

ggplot(my.data) + geom_histogram(aes(x=V1,fill=as.factor(V2))) + facet_grid( V2~.)

But I'd like the y-axes to overlap.

Answer 1

require(ggplot2)
require(plyr)

my.data <- as.data.frame(rbind( cbind( rnorm(1e3), 1) , cbind(     rnorm(1e3)+2, 2), cbind( rnorm(1e3)+3, 3), cbind( rnorm(1e3)+4, 4)))
my.data$V2=as.factor(my.data$V2)

calculate the density depending on V2

res <- dlply(my.data, .(V2), function(x) density(x$V1))
dd <- ldply(res, function(z){
  data.frame(Values = z[["x"]], 
             V1_density = z[["y"]],
             V1_count = z[["y"]]*z[["n"]])
})

add an offset depending on V2

dd$offest=-as.numeric(dd$V2)*0.2 # adapt the 0.2 value as you need
dd$V1_density_offest=dd$V1_density+dd$offest

and plot

ggplot(dd, aes(Values, V1_density_offest, color=V2)) + 
  geom_line()+
  geom_ribbon(aes(Values, ymin=offest,ymax=V1_density_offest,     fill=V2),alpha=0.3)+
  scale_y_continuous(breaks=NULL)

Answer 2

densityplot() from bioconductor flowViz package is one option for stacked densities.

from: http://www.bioconductor.org/packages/release/bioc/manuals/flowViz/man/flowViz.pdf :

For flowSets the idea is to horizontally stack plots of density estimates for all frames in the flowSet for one or several flow parameters. In the latter case, each parameter will be plotted in a separate panel, i.e., we implicitely condition on parameters.

you can see example visuals here: http://www.bioconductor.org/packages/release/bioc/vignettes/flowViz/inst/doc/filters.html

source("http://bioconductor.org/biocLite.R")
biocLite("flowViz")

Answer 3

Using the ggridges package:

ggplot(my.data, aes(x = V1, y = factor(V2), fill = factor(V2), color = factor(V2))) +
  geom_density_ridges(alpha = 0.5)

Answer 4

I think it's going to be difficult to get ggplot to offset the histograms like that. At least with faceting it makes new panels, and really, this transformation makes the y-axis meaningless. (The value is in the comparison from row to row). Here's one attempt at using base graphics to try to accomplish a similar thing.

#plotting function
plotoffsethists <- function(vals, groups, freq=F, overlap=.25, alpha=.75, colors=apply(floor(rbind(col2rgb(scales:::hue_pal(h = c(0, 360) + 15, c = 100, l = 65)(nlevels(groups))),alpha=alpha*255)),2,function(x) {paste0("#",paste(sprintf("%02X",x),collapse=""))}), ...) {
    print(colors)
    if (!is.factor(groups)) {
        groups<-factor(groups)
    }
    offsethist <- function (x, col = NULL, offset=0, freq=F, ...) {
        y <- if (freq) y <- x$counts
        else 
            x$density
        nB <- length(x$breaks)
        rect(x$breaks[-nB], 0+offset, x$breaks[-1L], y+offset, col = col, ...)
    }

     hh<-tapply(vals, groups, hist, plot=F)

    ymax<-if(freq)
        sapply(hh, function(x) max(x$counts))
    else
        sapply(hh, function(x) max(x$density))
    offset<-(mean(ymax)*overlap) * (length(ymax)-1):0
    ylim<-range(c(0,ymax+offset))
    xlim<-range(sapply(hh, function(x) range(x$breaks)))
    plot.new()
    plot.window(xlim, ylim, "")
    box()
    axis(1)

    Map(offsethist, hh, colors, offset, freq=freq, ...)
    invisible(hh)
}

#sample call
par(mar=c(3,1,1,1)+.1)
plotoffsethists(my.data$V1, factor(my.data$V2), overlap=.25)

Answer 5

Complementing Axeman's answer, you can add the option stat="binline" to the geom_density_ridges geom. This results in a histogram like plot, instead of a density line.

library(ggplot2)
library(ggridges)

my.data <- as.data.frame(rbind( cbind( rnorm(1e3), 1) , 
                                cbind( rnorm(1e3)+2, 2), 
                                cbind( rnorm(1e3)+3, 3), 
                                cbind( rnorm(1e3)+4, 4)))
my.data$V2 <- as.factor(my.data$V2)
ggplot(my.data, aes(x=V1, y=factor(V2),  fill=factor(V2))) +
      geom_density_ridges(alpha=0.6, stat="binline", bins=30)

Resulting image: