Hi everyone,
I’m currently using IsoformSwitchAnalyzeR to analyze an EMT kinetic study (3 timepoints) using PacBio Long Reads (quantified with IsoQuant).
I am encountering a surprising result: from an initial set of ~118,000 isoforms, the tool only identifies about 20 switching isoforms. This seems underestimated, especially since previous studies using short-read data on similar models usually identify hundreds of switches.
I have already tried to relax the filters (dIF > 0.1 and even 0.05) and adjusted the preFilter() parameters, but the number of significant results remains very low. I suspect the issue might be related to the p-value adjustment (FDR) being too stringent for this specific dataset, or perhaps a lack of power in the isoformSwitchTestDEXSeq().
Has anyone experienced this "underestimation" with Long Reads? Should I look into the dispersion estimation or is there a specific way to handle FDR for long-read counts in the package?
Thanks for your help
•
written
by
bendakiiracha17 •
0 answers
No answers yet.
Log in to answer this question.
The FDR is the False Discovery Rate - as a proceedure, it is neither too stringent, or insufficiently stringent depending on the data, it is what it is - the proportion of discoveries that can be expected to be false using a given p-value threshold. It does assume that the p-values are correctly estimated though. You might want to look at the distribution of unadjusted p-values though - it should be uniform with the exception of a excess of p-values at the very low end.
If you have previous short read data on similar models, I suggest that you find some positive controls and look at what is happening to them in your short read data at the level of raw counts.