UpSetR
UpSetR copied to clipboard
hms-dbmi
Exclusive vs inclusive intersections option
If I'm not mistaken there's no setting to specify in a plot whether intersections should be inclusive or exclusive - i.e. whether the value for AB is a count of all observations of AB, or if it excludes for instance ABC, ABD, etc, which have their own bars. UpsetR implements the latter, e.g. AB = 1 in following plot:
x = data.frame(A=1, B=c(0,1,1), C=c(0,0,1))
upset(x, sets = names(x))
Am I wrong and there is a way to get a value of 2 for AB with the above dataset (and 3 for A)?
If not, this seems to me a problem for 2 reasons. Firstly there's the obvious risk that some readers will interpret AB as inclusive. Secondly, for datasets featuring a large number of categories (or where the long tail (..AX) might be cut off by the nintersects argument) it's actually very difficult to gain from the plot even an approximate sense of the actual total intersection between A and B, let alone a figure. Doesn't that rather defeat the point of the whole approach?
For an example of a long dataset:
x = read.csv('https://gist.githubusercontent.com/diplodata/b2d6610a6e43fddfccea9a8a4821a154/raw/5d795f82f59e6571b78e99f8612edcbb7e5886b8/test.csv')
Take AD as an example. It's almost impossible from the plot to tell the intersection. sum(x$D & x$A) reports 21. But there's no way you could figure that out from looking at the graph.
upset(x, sets = names(x))
In my opinion, this is a major problem. The true functionality might be explained somewhere in the paper or elsewhere, but quickly looking at these graphs, I feel most people will expect the plot to be "inclusive". I would like to use these graphs to get away from having to do any mental math (which is the problem with Venn diagrams in the first place). I would prefer to look at the graph and immediately determine the inclusive overlap between any number of groups. I'm not sure it's possible, but I would like to request an option to make the graphs display inclusive overlaps. Thanks for a very cool approach!
Nice to see I'm not alone! Alas it's starting to look like UpSetR may be a dead project - a real shame if so.
Hi!
Is there any update on this issue? I am also interested in applying the "inclusive" version and would like to know if you found a way to do it.
I transitioned to using a different package, called ggupset. I asked how to solve this problem on their GitHub issues page and the developer provided a nice reproducible example. This might be helpful to someone: https://github.com/const-ae/ggupset/issues/3