DEXSeq_comparison.R -- questionable dpsi calls from events with psi=NA

[SolKatzman]

junctionCounts version 1.0.0

This is probably the proximate cause of item #21

In my previous sample-pairwise comparisons of junctionCounts psi calls, I have always excluded any events for which either of the samples had avg_psi value "NA"

I have gone back to extract and count such events for the set of samples that I have used as input to DEXSeq_comparison.R

At the following URL, I have created files that have these counts for each of these 8 event-types: (AL AF MX A5 A3 SE MS RI)

psi=NA stats

Each file has the counts of distinct events (counted with "sort | uniq | wc -l" ) for:
a) the 6 individual samples
b) the combined samples for each condition
c) all of the combined sample

It turns out that there is a lot of overlap of such events among the samples in out dataset.

Here are the item (c) counts for the 8 event-types:

A3 all samples psi=NA distinct events count: 20237
A5 all samples psi=NA distinct events count: 11545
AF all samples psi=NA distinct events count: 112803
AL all samples psi=NA distinct events count: 70922
MS all samples psi=NA distinct events count: 53058
MX all samples psi=NA distinct events count: 24739
RI all samples psi=NA distinct events count: 24806
SE all samples psi=NA distinct events count: 67085

I guess it is an interesting question as to why there are so many such events in our dataset.

In any case, to workaround the DEXSeq issue where it apparently does not exclude these events, I can filter them out myself before submitting them to DEXSeq.R

Sol Katzman
UC Santa Cruz Genomics Institute