ggplot2 & stat_ellipse: Draw ellipses around multiple groups of points

Question

This might be a simple one, but I'm trying to draw ellipses around my treatments on my PCoA plot.

My data frame (sc) is:

             MDS1        MDS2 Treatment
X1xF1 -0.19736183 -0.24299825   1xFlood
X1xF2 -0.17409568 -0.29727596   1xFlood
X1xF3 -0.15272444 -0.28553837   1xFlood
S1    -0.06643271  0.47049959     Start
S2    -0.15143350  0.31152966     Start
S3    -0.26156297  0.12296849     Start
X3xF1  0.29840827  0.04581617  3xFloods
X3xF2  0.50503749 -0.07011503  3xFloods
X3xF3  0.20016537 -0.05488630  3xFloods

and my code is:

ggplot(data=sc,(aes(x=MDS1,y=MDS2,colour = Treatment)))+geom_point(size=3)+
  ggtitle("PCoA of samples at 'class' level(method='Bray')\n",sep=''))+
  theme_bw()+guides(colour = guide_legend(override.aes = list(size=3)))+
  stat_ellipse()

It plots the PCoA okay up until stat_ellipse(). I've tried it with various parameters and at best I can get one ellipse for the whole plot (although I can't seem to reproduce that now).

What I'm after is three CI ellipses for the three treatments, coloured the same as the treatments. Any help would be very appreciated!

Thanks.

Solution

There is no stat_ellipse(...) in the ggplot package, so you must have retreived it from somewhere else. Care to share?? There are at least two versions that I am aware of, here, and here. Neither of these seem to work with your dataset, which is odd because both have worked with other datasets.

I finally fell back on the option of generating the ellipses externally to ggplot, which is not that difficult really.

library(ggplot2)
library(ellipse)
centroids <- aggregate(cbind(MDS1,MDS2)~Treatment,sc,mean)
conf.rgn  <- do.call(rbind,lapply(unique(sc$Treatment),function(t)
  data.frame(Treatment=as.character(t),
             ellipse(cov(sc[sc$Treatment==t,1:2]),
                     centre=as.matrix(centroids[t,2:3]),
                     level=0.95),
             stringsAsFactors=FALSE)))

ggplot(data=sc,(aes(x=MDS1,y=MDS2,colour = Treatment)))+
  geom_point(size=3)+
  geom_path(data=conf.rgn)+
  ggtitle(paste("PCoA of samples at 'class' level(method='Bray')\n",sep=''))+
  theme_bw()+
  guides(colour = guide_legend(override.aes = list(size=3)))

Comments

[Vlo]

The main issue is that stat_ellipse() is a wrapper function for car::ellipse() and geom_path(), and its default parameters trip up when there are a limited number of data pints.

[RB88]

Thanks jlhoward. stat_ellipse() is from devtools, sorry I forgot to mention it. I've tried using your code, but the ellipses aren't appearing, it just generates the normal plot, and I can't figure out why...

[jlhoward]

Try restarting R, then copy/paste the code exactly.

[RB88]

That didn't work unfortunately. When I type conf.rgn after running your function, I get a 300 x 3 data frame with col 1 as treatment (100 x each treatment), and col2 & 3 as MDS1 & MDS2. These final two columns are 'NA', so I think that's where the problem lies.

[jlhoward]

Hard to tell what's going on here. Do you get any error messages or warnings? Does your code to produce sc load any packages? If you load the car package, or if you load a package that requires car, that will mask the ellipse(...) function. But this should result in error messages.

[RB88]

I've got it working! I had as.character(sc$Treatment) when I was setting my factors, and removing that solved it. Now I just need to reorder the legend and change the legend symbols, and I'm sorted. Thanks for your help!