I was simply wondering whether it was possible to get an estimate for the differential usage of each isoform instead of each gene? This would make downstream analysis a lot easier since one would know exactly which isoforms were differentially used instead of just knowing that some (of the potentially many) isoforms within this gene were changing.
It would basically correspond to getting a p-value for each isoform proportion/fraction.
I have been working on implementing the isoform-level analysis which are based on the beta-binomial distribution.
Additionally, there will be a full regression framework available for gene-level and feature-level analysis allowing for modeling more complicated designs.
All these features will be available with the new Bioconductor release 3.5.
Currently, the development version of these implementations is available on DRIMSeq github https://github.com/markrobinsonuzh/DRIMSeq, but I suggest to wait until they are pushed to the devel version of Bioconductor since they are updated very intensively.
That sounds extremely usefull - both the feature level test but also the complex designs! Looking forward to try it!