PSI filter recommendation

[Newuser-lab]

Hi,
In your Paper you recommend to filter for:

• Average read count > 10 in both sample groups
• Filter events with average PSI between 0.05 and 0.95 in both sample groups
• FDR <= 0.1
• Abs(deltaPSI) >= 0.05

I want to prepare my data for further downstream analysis (e.g. motif enrichment analysis around splice sites, RBP sites etc.). I was thinking to filter my data as following:

1. Read count (IJC + SJC) for each replicate above 10 in both sample groups
2. Average IncLevel between 0.05 and 0.95 in both sample groups
3. Abs(IncLevelDifference) > 0.1
4. FDR < 0.1

In issue #440, you say that the PSI filter (point 2) does not remove anything that wouldn’t be removed by filter 3 and that she should remove the filter. But the filter does remove more than filter 3 would remove alone. You exclude all events that are fully excluded or included in one sample (IncLevel = 1 or 0) but alternatively spliced in the other sample, right?
For point 2, do you recommend to use the filter to prepare the data for downstream analysis in any case? Why do you recommend this filter?

Many thanks for your input!

[EricKutschera]

From #440

2. filter out events with average PSI value <0.05 or >0.95 in both sample groups

4. abs(deltaPSI) >= 0.05

filter 2 doesn't remove anything that wouldn't already be removed by filter 4

Here's the implementation of that filter: https://github.com/Xinglab/rmats-turbo-tutorial/blob/2038e61a1db9e7091c3d90901b7eaee7ae9425b6/scripts/rmats_filtering.py#L45

Here you have a different wording of the filter

Filter events with average PSI between 0.05 and 0.95 in both sample groups

Then the question is about a case where the average PSI is 1.0 in one group and something like 0.9 in the other group. With the wording from #440 the event passes the filter because the PSI is only >0.95 in one of the groups not "in both sample groups". With the wording in this issue the event is filtered out because one of the groups has a PSI >0.95

I think the set of filters was initially based on #183 (comment) but then modified. The original average PSI filter was used in a situation without 2 groups, but in the 2 group setting delta PSI can be used instead