Hi,

I have been following the DTU workflow which is linked to the package rnaseqDTU and the paper: "Swimming downstream: statistical analysis of differential transcript usage following Salmon" https://f1000research.com/articles/7-952/v3

It is mentioned that, in favor of controlling the FDR of DRIMSeq one should apply a post-Hoc filter to exclude "transcripts with small SD of the per-sample proportions". This was not suggested to be done for results of DEXSeq and its workflow which is described in the same paper. The reason is that DEXSeq seemed to "be closer to controlling its FDR".

I wonder if this post-Hoc filter would be redundant if one is only interested in the intersection of DRIMSeq and DEXSeq results. As those FPs that were called by DRIMSeq would in the best case not be present in DEXSeq. (Having in mind that the intersection of results from both tools might lack some DTUs that were found by DRIMseq being more sensitve).

Thanks in advance

Fiona

I suppose that you could skip the filter then. The filter is just a suggestion (and I should said that it's only coming from my benchmark, not from the package authors). I think of it as analogous to post-hoc filtering of genes with small LFC. This is common practice and easy to do, although in theory the minimum effect size could also be incorporated into the testing procedure somehow (e.g. TREAT and lfcThreshold in edgeR/limma and DESeq2 respectively).