Hello everyone,

I am getting some surprising behaviour out of the ChIPpeakAnno R package.  The question I’m trying to ask is ‘Which features intersect between samples ‘WT312’, ‘WT340’, and ‘WT354’, that also intersect with the intersection between samples ‘WT313’, ‘WT341’, and ‘WT355’. To visualize this, I am using makeVennDiagram to create two three-way venn diagrams: one for samples ‘WT312’, ‘WT340’, and ‘WT354’, and one for samples ‘WT313’, ‘WT341’, and ‘WT355’.  Then I use findOverlappingPeaks to make an intersection of WT312, WT340, and WT354, which I call WT312_WT340_WT354.  Separately, I also use findOverlappingPeaks to make an intersection of WT313, WT341, and WT355, which I call WT313_WT341_WT355.  Then I take these two three-way intersected lists and pass those back through to makeVennDiagrams.  The issue is that the total number of elements contained within the three-way venn diagrams for WT312-WT340-WT354 and WT313-WT341-WT355 do not sum up to the total number of elements from the two-way venn diagram.  I’m attaching three PNG files to show what I’m talking about.  Specifically, the number of elements that are found in WT312-WT340-WT354 in the two-way venn diagram exceeds the number of elements found to intersect between WT312-WT340-WT354 in the three-way venn diagram.  The same basic story exists for WT313-WT341-WT355.

I can’t understand what would cause this behaviour, but it is a bit concerning for me.  The intersection between the sets WT312-WT340-WT354 and WT313_WT341_WT355 should exactly sum to the intersection of the set WT312-WT340-WT354 and the the intersection of set WT313-WT341-WT355.

I'm pasting image links so you can see what I'm referring to.

487+395 = 882, not 852 

and

292+395 = 687, not 674 

Thanks,
Brad Davis

Hi Brad,

Thank you for your message. This question is very typical one for calculating intersection of peaks. As you know a peak is not a point, it is a range of continuous points. If we consider intersection of set A (1-2, 4-5, 7-9) with set B (2-8), how many intersect peaks should we got? This is a question. If I say 1, but setA has three peaks. If I say 3 but setB has only one peak. In ChIPpeakAnno (release version), we are using the minimal number for intersect peaks. That is to say we are using 1 in this case. This is way you will get smaller number than it looks like. 

I know this issue make people uncomfortable. I am thinking how to resolve this problem. I am thinking to build a new method to show the two numbers in the intersection part of venn diagram. I will let you know once it is done.

Hi Brad,

Please have a try the newest devel version at http://pgfe.umassmed.edu/ChIPpeakAnno/. Try makeVennDiagram with connectedPeaks="keepAll"

Hello Jianhong,

Thank you for your reply.  I understand your example and your point.

Cheers,

Brad

Hi Jianhong,

It seems like this behaviour should be related to the 'select' parameter option in findOverlappingPeaks command?  If so, shouldn't there be a way to generate identical venn diagrams by setting the 'select' command to a defined argument, such as 'left'?

Also, would it be possible to report the percentage (as opposed to the raw number) for each of the elements in the venn diagram? That would make it easier to compare between comparisons.