Hi All,
I have noticed a discrepancy when identifying differential binding sites using either edgeR or DeSeq2. I do not believe this is a normalization issue as I have checked both the box and MA plots using both raw and normalized counts and there is very little change.
I am looking at H3K27me3 ChIP-seq data between two different tumor subtypes (i.e BRAF vs RAS). When using edgeR many of the differential sites are in the RAS condition, but when using DeSeq2 many of the differential sites are in the BRAF condition. I have pasted my code below:
dba <- dba(sampleSheet = "file.csv", minOverlap = 2)
dba.count <- dba.count(dba, fragmentSize = 1000, minOverlap = 2)
dba.contrast <- dba.contrast(dba.count, categories = DBA_TISSUE)
#DESEQ2
dba.analyze <- dba.analyze(dba.contrast, method = DBA_DESEQ2, bSubControl = TRUE, bFullLibrarySize = TRUE)
~1800 differential sites in BRAF and ~ 100 differential sites in RAS
#EDGER
dba.analyze <- dba.analyze(dba.contrast, method = DBA_EDGER, bSubControl = TRUE, bFullLibrarySize = TRUE)
~200 differential sites in BRAF and ~ 800 differential sites in RAS
I have also tried setting the method in dba.count, using that for the contrast, and subsequently using the same method in dba.analyze (i.e. all DBA_DESEQ2) but am still getting the same result.
Also, while I check UCSC/IGV to determine the proper binding patterns, this is still rather troubling to me as I like diffbind. Any ideas to what could be going on?
Thanks very much for any help!
Chris
