Alternative splicing: explanation about delta psi values

Question

[user31888]

I am a bit confused.

I compared differentially expressed spliced transcripts between a sample 1 and a sample 2 using MISO algorithm.

The final (filtered) result table notably mentions the following info (this is an example for a transcript with 3 isoforms):

. delta PSI: -0.67,0.19,-0.87

. isoforms: isoform1,isoform2,isoform3 (isoform 1 is the inclusive form (with all exons); isoforms 2 and 3 are spliced)

According to MISO paper:

. PSI is the fraction of the inclusive transcript in a sample (i.e. not spliced if I understand well)

. delta PSI = PSI(sample1) - PSI(sample2)

Thus, for example:

. delta PSI(isoform1) = -0.67 means there are more inclusive transcript in sample 2 than sample 1,

. delta PSI(isoform2) = 0.19 means there are more inclusive transcript in sample 1 than sample 2, but how is it related to isoform2 which is known to be spliced? Does MISO report only the spliced isoforms in its output?

[aarin.jones]

Can I ask how you filtered your output when your sample has more than two isoforms? I'm aware that the documentation allows you to filter using the filter_events script, but thats only for two isoform events. You can filter on bayes-factor on >2 isoform events but this still outputs an note/error for me: "filter_events: filtering MISO pairwise comparison output.

Note: This utility is only works on MISO output for two-isoform event annotations.

Error: filter_events.py is only defined for MISO output on two-isoform alternative events. Found a non-two isoform event: ENSMUSG00000032469"

Answer 1

[Kristoffer Vitting-Seerup]

I must admit I'm also a bit confused. Note sure how to interpret PSI values for isoforms (instead of an exon) since it have to be different than Isoform Fractions (as they are not summing to zero).

You could take a look at IsoformSwitchAnalyzeR to see whether that does what you want.

[user31888]

I agree. It does not make any sense for isoform-centric analysis. I will definetely try your program. Does it require technical/biological replicates?

[Kristoffer Vitting-Seerup]

Yes (like everything else where you want to do statistics). It can however be run without but then the results are not trustworthy.