Problems running paired model in rMATS v4.1.1

[Alexander Greben]

Hi, I've been successfully running rMATS turbo v4.1.1, but I can't get it to work with the paired stats model. I'm working on a computing cluster and I had some issues getting the original dependencies right, I'm wondering if my issue has something to do with R dependencies again. It looks like rMATS isn't passing a properly-formatted object to PAIRADISE.

This is the error I get:

error in paired model
Loading required package: nloptr

Error in PAIRADISE::pairadise(data_frame, numCluster = number_of_threads) :

is(pdat, "PDseDataSet") is not TRUE

Calls: <Anonymous> -> stopifnot

Execution halted

It also says this:

paste: outdirname/tmp/JC_SE/rMATS_result_FDR.txt: No such file or directory

Thanks for any help you can offer.

Alexander

[kutsc...@gmail.com]

The error `is(pdat, "PDseDataSet") is not TRUE` will happen if you have an older version of PAIRADISE installed. rMATS is compatible with the latest version of PAIRADISE from github: https://github.com/Xinglab/PAIRADISE . The version on bioconductor is not up to date with the latest changes on github

The build script (`./build_rmats`) will only install PAIRADISE if there is no other version of PAIRADISE already installed: https://github.com/Xinglab/rmats-turbo/blob/v4.1.1/build_rmats#L77
https://github.com/Xinglab/rmats-turbo/blob/v4.1.1/install_r_deps.R#L24

You can try uninstalling PAIRADISE and reinstalling it from github

Eric

[Alexander Greben]

Hi Eric,

Thank you very much, that seems to have resolved the problem. It runs just fine now.

Interestingly, I'm seeing consistently fewer significant events when I run the paired model. For one dataset (3 paired samples), the results are as follows:

This is a dataset for which differential expression (by DESeq2) runs much better in paired mode. I see the same thing in another dataset as well. Is the paired analysis less sensitive, does it require more read depth per sample, or something else I'm not thinking of? The other optional settings I'm running are --variable-read-length and --allow-clipping.

Thanks,

Alexander

[kutsc...@gmail.com]

I'm not sure how the results of the different models are expected to compare. One thing you could do is manually check some of the rows in the *.MATS.JC.txt files where the models disagree about whether an event is significant and also rows where the models agree. Maybe there will be a pattern such as the paired model requiring more reads as you suggested

You could also try reaching out to the developer of PAIRADISE (Levon): https://github.com/Xinglab/PAIRADISE#contact

[Travel Well]

Hi Alexander,

I wonder if you have figure out the problem of paired model. I am having the same problem as you did - much lesser significant events from paired comparisons.

Thanks!