[rMATS] Number of AS Events from one sample

Question

[vin.darb]

Hi !

I have run rMATS to see if there any genes that are differently spliced between my two conditions (three replications) and I have now several genes to study, but I'd like to know if there is a way to access to the number of AS events detected into a single bam file ? Indeed I would like to see now if a particular type of event is more represented in my mutants, but I believe that rMATS study events by comparing two conditions.

STAR output a bed file with spliced junctions detected from each alignement, (file SJ.out.db), is there a way to access from this file, Is there way from the lines of this file to know what type of event this refers?

Thank's by advance

Answer 1

[yxing]

There is an easy trick for this. You can copy your file on any given sample of interest to another file with a different file name, and run a fake one vs one comparison using rMATS. Then rMATS would be able to identify and report the number of events corresponding to different types of AS patterns in your sample.

[Kristoffer Vitting-Seerup]

So you would be comparing a specific sample to itself?

[vin.darb]

Thank's for the answer

I follow you're technique, so I have only to take into account this information from rMATS:

Done processing each gene from dictionary to compile AS events Found 13475 exon skipping events Found 879 exon MX events Found 24992 alt SS events There are 14836 alt 3 SS events and 10156 alt 5 SS events. Found 24543 RI events

I will do this with every sample to see if I have more events in my mutants

[Kristoffer Vitting-Seerup]

Just remember this will be VERY sensitive to the expression cutoffs applied to the data. Furthermore most events will probably be found in both conditions but will just be used more/less in one condition (hence the statistical analysis in the first place).

Answer 2

[Kristoffer Vitting-Seerup]

If you are interested in a genome-wide summary of what the splice changes are I would recommend using IsoformSwitchAnalyzeR. Although it performs a transcript based analysis (as discussed previously here) I recent made an extension directly enabling statistical analysis such genome wide changes (both for alternative splicing and isoform switch consequences). You can see examples of what IsoformSwitchAnalyzeR can do in this section of the vignette and read the preprint about the extension here.

[vin.darb]

Thank's I will have a look on this package

I have another question, I have perform a DTU (Differential transcript usage) analyse using RATs package, in order to find Isoform switch between my two conditions. With this tools I find 1000 DTU genes which are almost all include in the 3000 alternatively spliced genes list that I generated with rMATS.

I thought I'd get the same list between the two methods but that would mean that alternative splicing would not necessarily imply Isoform switch ?

Or it's just that RATs is more stringent than rMATS ?

Thank's by advance ?

[Kristoffer Vitting-Seerup]

Comparing rMATS and RATs is a bit problematic because they approach the problem very differently. rMATS only considers each splice site one at the time. Compared to a transcript level analysis the analysis of a specific site is a much easier problem and thereby rMATS is probably more sensitive (captures more). But on the other and a transcript level analysis is much easier to interpret in a biological context. So it is a tradeoff between sensitivity and interpretability.

Another thing to remember is to consider effect size - it might be that many of the detected events have such a small effect (e.g 3% more/less usage in one condition). Did you use any effect size cutoff to get to the 1000 and 3000 genes?

Lastly RATs is known to be a quite conservative compared to other DTU tools (see this recent benchmark (starting at figure 6t)). Love et al speculate it is due to the many build in filters in RATs. In any case they show DEXSeq is a strong competitor which is why IsoformSwitchAnalyzeR uses DEXSeq instead (also due to runtime). Love et al also highlight the importance of filtering out low-count features before doing the analysis. Did you do any filtering before the statistical analysis?

Cheers Kristoffer

[vin.darb]

for both tools I filter out genes that have fewer than 10 average expressed transcripts between my 6 samples (3 samples by condition)

RATs use a isoform switch change ratio of 0.2 , so with rMATS I use a cutoff of dPSI of 0.2 too

With DEXSEQ I found a little bit more genes than rMATS, and 70% of the genes found by rMATS overlap with genes founds by DEXSEQ

[Kristoffer Vitting-Seerup]

Nice with the effect size cutoffs! When you mention DEXSeq do yo mean running it on exons (where are comparison to rMATS would be good) or running it on transcripts (as done in the Love et al article I mention above) where a comparison to RATs would be more appropriate?

Answer 3

[Charles Warden]

I hope you don't consider this off-topic: however, if you have replicates (which I strongly recommend), I personally would prefer to use QoRTS + JunctionSeq (although that analyzes each exon/junction separately with visualization, rather than having a splicing percentage)

[vin.darb]

I had looked at JunctionSEQ but I had seen that many people were using rMATS lately so I have choose rMATS, but I can compare both softwares to the the differences