I am reproducing a differential exon usage analysis using DEXSeq that was done on a previous version of the human reference genome (not sure which) with respect to the one I am using now (ensembl release 111).

The previous version reports 36 exons for my gene of interest (myo6), while the one I am using reports 73 exons. The exons I am looking for in silico were previously validated in the lab.

I am performing a DEXSeq standard analysis and I have 3 conditions, where 1 is the control and the other 2 are treatments. I performed the dexseq complete pipeline separating the two cases:

1. treatment 1 vs control
2. treatment 2 vs control And the output is then composed by two tables. However, the results are no longer significant for the exons of interest, while they were in the previous analysis (with the reference genome containing less exons).

Then I tried performing the dexseq analysis by creating a unique model using all 3 conditions and specifying in the fold change computation denominator = control. In this way the exons of interest are significant again. However, this method outputs 30k significant events versus the 500 outputted by the previous method.

I am wondering whether the different number of exons in the reference affects the DEU analysis of dexseq and if there is any possible solutions.

Likely the new annotation that you are using has more annotated isoforms, and thus the exons are splitted into more disjoint exonic bins. I'd expect some differences in the output, but they should not be completely discordant. Something I'd check is whether you are using the same full and reduced models that were used in the old analysis.