How To Calculate PSI with JunctionSeq

[xysj1989]

Dear developers,

I have tried JunctionSeq in my project (RNA-Seq-based analysis of alternative splicing during plants development ). JunctionSeq is really a useful tool with powerful visualisation function. I have 3 questions below.

1. Is it possible to calculate PSI (percent spliced in) for identified alternative isoforms through JunctionSeq ?
2. Have you already compared the results from JunctionSeq with other tools like rMATS or SUPPA2?
3. How to join JunctionSeq and rMATS/SUPPA2 analysis together in an alternative splicing analysis project? Because, JunctionSeq can currently provide beautiful images of Splicing Event, while the other tools can provide PSI calculation.

Best Wishes!

[hartleys]

Unfortunately there's no real way to reliably calculate percent spliced in using JunctionSeq's methods. Sequence- and locus-specific bias is very strong in RNA-Seq, meaning you can't directly compare the coverage depth of locus A to the coverage depth of locus B. All we can do is compare the difference in coverage depth of locus A between two conditions, relative to the difference in coverage depth of locus B between two conditions (in JunctionSeq locus A is the current junction and locus B is the rest of the gene as a whole). You always have to compare each locus to itself though.

That's my opinion anyways. rMATS just ignores the problem and reports the comparison under the assumption that no such biases exist. I don't think I'd be comfortable making JunctionSeq do that...

To see an example of the coverage bias, look at the coverage depth plots shown in fig2 of my rat pineal gland paper:
https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0163590

Even just across a single exon the coverage varies quite a bit. In that first exon the coverage varies by a factor of 2 between the start and end. This can't be due to actual differentials, since the changes happen along a single unbroken exon. It's just an artifact of how we sequence and process the data.

So if you just compare depths directly between loci you can expect results to be off by a factor of 2-4. That seems like too much.