Confidence that Alternative Splicing is NOT happening

[Germwise]

Hey all. First post, sorry if it's a stupid question.

To make a long story short:

I've inherited a project in our lab working on a putative alternative splicing event of a gene. Instead of messing with wet bench experiments I decided to analyze some RNA-Seq data to determine how prevalent this isoform may be.

I've combined something like 50 SRA files and done tophat alignment. I've found every gene junction except for the one that was predicted to be alternatively spliced! The problem is that this gene is expressed at very low levels and I may not be seeing the AS version because of overall reads.

Is there something I can use that will count the total reads aligning to the gene or the number of times the other junctions have been found and give me a confidence or p-value that the AS event does NOT exist?

Essentially I need to make a mathematical decision on whether it is worth continuing this project or to maybe conduct another RNA-Seq to gain more confidence.

Thanks in advance!

Germwise

[frozenlyse]

If you already know what splice junction(s) you are interested in interrogating, and already have the RNA sample(s) you wish to test, why not just do the qPCR? It's super quick and easy, and more sensitive than RNA-seq.

[cedance]

I don't quite understand why do RNA-Seq to study AS of 1 gene? As frozenlyse says, wouldn't a q(RT)-PCR be the best way to find out isoform abundance?

[Germwise]

the mRNA-seq experiments were downloaded so they were free and I wanted to complement my qRT PCR results showing that this isoform existed.

However, it was extremely surprising that the RNA-Seq data didn't find the splice junction.

Anyone have an idea on how to do the statistics on it so that I can see how confident I can be in these results?

[cedance]

If you are looking for an exon skipping AS event, for example, then one way to go about would be to find all the isoforms of this gene from the data you have (cufflinks/cuffdiff) and find the inclusion level of all exons for this gene.
From your lab experiments what AS and how many isoforms were you able to see, any idea? Did they also have intron retention events?

[Germwise]

The AS of this gene is intron retention (and this is the majority form)

*Tophat does not find the junction of the intron

*When I input the predicted gene models into cufflinks, it gives me 3-10pct value for the form with the properly spliced intron. However, without finding the junction, I'm not sure the math behind its prediction.

I'm attaching a picture of part of my alignment showing both splicing gene models at the bottom. The intron retention form is easily found but no splicing junction.

Help!

I'll add that my experimental data suggests just these 2 forms though these results may be an artifact.

[dpryan]

I'm not familiar with a tool to do this off-hand, but you could create a monte-carlo simulation that would let you know the probability of finding the junction or not given the data (and probably a few assumptions). Having said that, it wouldn't be trivial to write and wouldn't be worth the effort unless there's some super compelling reason to suspect that this splicing event takes place (i.e., a reviewer could reasonably ask for it).

[Germwise]

dpryan-

There is no reviewer as of yet, but I since I have data for and against AS, I want to know how much weight I should place on these RNA-Seq results. Basically, it comes down to a decision on whether we will pursue this further.

[frozenlyse]

I would test that tophat actually *can* map reads across that splice junction - simulate reads (of the same length as the datasets you have downloaded) across the whole of the two isoforms, and see how well they map back to preclude some sort of mapping bias.

Unless the public data you have downloaded is the same tissue you are interested in, it's never going to be able to prove that a certain AS event does *not* happen. Have you done RT-PCR with a primer straddling that spice junction?

[rskr]

It depends on you being able to calculate the expected insert size of the RNA-seq data.