Hi I am using Diffbind to report the DBS (I know that SICER has already a tool for doing so, but I have a huge dataset with different treatments and I just want to try with DiffBind)
In any case, have you ever tried it? Even if the results that I get are good and consistent with other peak callers, it is a bit weird to me to do not find any correlation among the samples in the occupancy correlation heatmap plot. I am using the test-W200-G600-FDR.01-island.bed files.
It's fairly normal not to see much correlation at the stage of having loaded the sample sheet, before counting reads (and sometimes even after counting). ChIP-seq can be pretty noisy, and in a straight occupancy analysis, even tiny noise peaks have the same weight as bigger ones. If you still see weak clustering after the dba.analyze step, then there's a problem.
If you post an example heat map, I can tell you whether it looks similar to what I usually see.
Ok! thanks. It is only before counting the reads, after is ok (check the occupancy heatmap bellow). I was a bit concerned because I used other peak callers for the same dataset and I could see some correlation in the occupancy heatmap, but not when I used SICER. I was just wondering if that was because I was using the wrong bed file , but I got more or less the same results in the GO associations, so I guess that it is ok (test-W200-G600-FDR.01-island.bed).
I also thought that it could be because the bed file was a bit different , but it seems that is not the case. With some peak callers I needed to change a bit the output file in order to get a correct input for DiffBind (MUSIC for example)
There's something very wrong with your heat map. There's no way every sample could be exactly the same distance apart from every other sample, which is what this shows (unless the data were artificially generated to create this pattern). I can't really imagine how you could produce this via DiffBind. Could you post (or email to me, if you prefer) your sample sheet and the top 50 or so lines of each peak file?
Just a clarification, I just realised that there is not need to change the MUSIC output bed file. I did because I saw that was a bit different from MACS output, but there is not need.
Hi,
There's something very wrong with your heat map. There's no way every sample could be exactly the same distance apart from every other sample, which is what this shows (unless the data were artificially generated to create this pattern). I can't really imagine how you could produce this via DiffBind. Could you post (or email to me, if you prefer) your sample sheet and the top 50 or so lines of each peak file?
- Gord