Interpreting rMATS Summary Table and Event Ratios

[X L]

Hi rMATS users,

I have a question about interpreting the summary table. For the A3SS event type, I have SignificantEventsJC (2609), with SigEventsJCSample1HigherInclusion (1734) and SigEventsJCSample2HigherInclusion (875). My control dataset is sample 2, and my treatment dataset is sample 1. Does this mean that rMATS detected a total of 2609 significant events based on JC between my treatment and control samples, with 1734 events in the treatment favoring the usage of the 3′ splice site closer to the 5′ splice donor (leading to a longer downstream exon), and 875 events favoring the usage of the 3′ splice site further away from the 5′ splice donor (leading to a shorter downstream exon)?

Additionally, I’m wondering if the ratio between SigEventsJCSample1HigherInclusion and SigEventsJCSample2HigherInclusion matters. For example, in this case, I have 1734/875 = 1.98 for A3SS, but for my SE (skipped exon) event type, the ratio is 4955/5747 = 0.86. Does this imply that I should pay more attention to the A3SS events than to the SE events?

Thank you,
Xiao

[kutsc...@gmail.com]

SignificantEventsJC 2609 means that 2609 events were found as significant using JC based on (https://github.com/Xinglab/rmats-turbo/blob/v4.3.0/rMATS_P/summary.py) with a default of FDR 0.05

SigEventsJCSample1HigherInclusion 1734 means that 1734 of the 2609 had IncLevelDifference greater than 0. Since IncLevelDifference = average(IncLevel1) - average(IncLevel2), that means that sample group 1 had a higher inclusion level on average

From the README section about A3SS and A5SS coordinates:
> The inclusion form includes the long exon (longExonStart_0base, longExonEnd) instead of the short exon (shortES shortEE)

If the treatment dataset is sample group 1 then there are more significant A3SS events where the treatment uses the longer exon more than the control

The ratio of Sample1Higher to Sample2Higher could be interesting as a measure of how often the longer exon is used in your treatment dataset. The SE events could still be interesting even if there is no clear bias toward more inclusion or more skipping in your treatment dataset

Eric

[X L]

Hi Eric,

Thank you very much for your explanation!

I'd like to confirm my understanding: Are all rMATS events based on a binary competitive model of splice site selection? For example, in your reply, you mentioned that for events with IncLevelDifference > 0 and FDR < 0.05, and if the treatment dataset is sample group 1, then in such significant A3SS events, the treatment uses the longer exon more frequently than the control.

Does this mean that in such events, the proximal 3' splice site is favored (used more) while the distal 3' splice site is disfavored (used less)? In other words, is it correct to assume that when one splice site's usage increases (e.g., proximal 3' splice site), the other's usage must correspondingly decrease (e.g., distal 3' splice site) in a mutually exclusive manner?

Am I on the right track with this interpretation?

Best regards,
Xiao

[kutsc...@gmail.com]

Hi Xiao,

rMATS only considers two isoforms for each event. The inclusion level is a ratio of reads that support those isoforms. Here's the calculation: https://github.com/Xinglab/rmats-turbo/issues/349

The inclusion level is another name for PSI (percent spliced in). If the percent for the inclusion isoform goes up, then the percent for the skipping isoform goes down. The percent for the skipping isoform would be (1 - PSI)

For A3SS events, the two isoforms correspond to the two splice sites. The inclusion level increasing means that the proximal splice site is relatively used more and the distal splice site is relatively used less

Eric