I've been working through the DEXSeq vignette and noticed for the default summarizeOverlaps method of counting reads that some important parameters are missing, namely inter.feature. According to the method documentation, the default call to summarizeOverlaps with BamFile as input would set inter.feature to TRUE, however based on my understanding of the DEXSeq algorithm as described in the paper I don't think this is the desired behavior as it would not provide counts to reads spanning multiple features (i.e., exon bins).
I think the correct way of calling summarizeOverlaps is this (with mode and inter.feature specifically set):
se = summarizeOverlaps(
flattenedAnnotation, BamFileList(bamFiles), singleEnd=FALSE,
fragments=TRUE, ignore.strand=TRUE, inter.feature=FALSE, mode="Union")
It seems this was implicit in an older version of the vignette but was removed in a recent commit. Am I wrong here about the desired behavior and the correction was intentional? I think it's worth clarifying as not counting overlapping features would significantly impact the resulting counts as it's likely a large portion of reads would span multiple bins.
I've been working through the DEXSeq vignette and noticed for the default
summarizeOverlapsmethod of counting reads that some important parameters are missing, namelyinter.feature. According to the method documentation, the default call tosummarizeOverlapswithBamFileas input would setinter.featuretoTRUE, however based on my understanding of the DEXSeq algorithm as described in the paper I don't think this is the desired behavior as it would not provide counts to reads spanning multiple features (i.e., exon bins).I think the correct way of calling
summarizeOverlapsis this (withmodeandinter.featurespecifically set):It seems this was implicit in an older version of the vignette but was removed in a recent commit. Am I wrong here about the desired behavior and the correction was intentional? I think it's worth clarifying as not counting overlapping features would significantly impact the resulting counts as it's likely a large portion of reads would span multiple bins.