I am studying alternative splicing in some genes. The original data were analyzed a couple of years ago with DEXSeq (I ignore which version was used as the analysis was done by a previous colleague). To integrate the information about exon usage with infomration about differential isoform expression I run BitSeq and Cuffdidd as comparison. While the number and the pattern of differentially expressed isoforms are quite consistent between BitSeq and Cuffdiff the comparison with DEXSeq is quite puzzling. Only half of the isoforms detected by DEXSeq are significant also in BitSeq/Cuffdiff output and even more confusing some of them show an opposite pattern (for a specific exon which show a negative log2fold change between 2 conditions in DEXSeq the corresponding isoforms have a positive log2fold change between the same 2 conditions in BitSeq/Cuffdiff).
I can't think of a possible biological explanation for that as I would expect that at least one of the corresponding isoforms of a more expressed exons would also be more expressed. I was planning to repeat the DEXSeq analysis with the current version (1.14), can I expect results more compatible with the BitSeq/Cuffdiff output? Are you aware of similar inconguencies with the previous version of DEXSeq?
Also, to make the comparison with BitSeq (transcriptome-based) more reliable we are planning to align the reads to both genome and transcriptome for DEXSeq 1.14, is this a good strategy?