about psiPerEvent operation

[renzhonglu]

Dear Eduardo,
I have a question about psiPerEvent function. After using generateEvent function from GTF file to obain SE SS MX RI FL, 5 ioe files, should I use these ioe files as input into psiPerEvent function separately, or merging 5 ioe into 1 ioe file before psiPerEvent operation? I am not sure that SUPPA regards different alternative splicing patterns as irrelevant events.
If so, when using diffSplice, it aslo should be calculated p-value and p.adj value in 5 patterns separately. there are different number of significant local events in different patterns , so the multiple test correction p-values should be different. For example, an ES event has significant difference between 2 conditions after multiple test correction in ES pattern, but when merging all events of 5 patterns, this ES event may not be significant or may be with different p.adj values due to changes of number of total events.
I also notice that in your paper "Large-scale analysis of genome and transcriptome alterations in multiple umors unveils novel cancer-relevant splicing networks" Genome Research 2016, BH methods were used to correct for multiple testing. So, should I only use diffSplice to generate original p-value in 5 patterns, and then merge total events of 5 patterns to calculate p-adjust value using BH correction method, myself?

Than you very much!
Kind regards,
Elias

[EduEyras]

Hi Elias, For PSI calculation you can merge all the vents into one file. For diffSplice calculation it is better to do it with all together, so that you can do per gene correction using all events in each gene. In the paper Sebestyen et al. 2016 we were using a non-parametric test (Mann-Whitney) to compare two sample groups. Since we were doing this irrespective of the gene expression levels, we had to add some multiple test correction. That is why we used BH. If you use the empirical method with diffSplice, you can use the per-gene correction, which performs a correction per gene. This takes into account the fact that all events in the same gene are compared to the similar set of genes, but it would not be as harsh as doing BH with all events. You could also do BH if you wanted to be overly conservative. For BH you may also remove the events that were not considered for analysis because the low expression of the genes. This is how we did in Sebestyen et al. 2016. I hope this helps Best Eduardo

…

On Wed, 27 Mar 2019 at 21:08, renzhonglu ***@***.***> wrote: Dear Eduardo, I have a question about psiPerEvent function. After using generateEvent function from GTF file to obain SE SS MX RI FL, 5 ioe files, should I use these ioe files as input into psiPerEvent function separately, or merging 5 ioe into 1 ioe file before psiPerEvent operation? I am not sure that SUPPA regards different alternative splicing patterns as irrelevant events. If so, when using diffSplice, it aslo should be calculated p-value and p.adj value in 5 patterns separately. there are different number of significant local events in different patterns , so the multiple test correction p-values should be different. For example, an ES event has significant difference between 2 conditions after multiple test correction in ES pattern, but when merging all events of 5 patterns, this ES event may not be significant or may be with different p.adj values due to changes of number of total events. I also notice that in your paper "Large-scale analysis of genome and transcriptome alterations in multiple umors unveils novel cancer-relevant splicing networks" Genome Research 2016, BH methods were used to correct for multiple testing. So, should I only use diffSplice to generate original p-value in 5 patterns, and then merge total events of 5 patterns to calculate p-adjust value using BH correction method, myself? Than you very much! Kind regards, Elias — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#53>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMWVBxoFhmjDawK-EN0DlCOxERsFl1sCks5va0MygaJpZM4cNZpx> .

-- Prof. E Eyras EMBL Group Leader The John Curtin School of Medical Research Australian National University 131 Garran Road, Acton ACT 2601, Canberra, Australia https://github.com/comprna http://scholar.google.com/citations?user=LiojlGoAAAAJ http://www.researcherid.com/rid/L-1053-2014

[renzhonglu]

Hi Eduardo

Thank you for your reply.

I will merge all the events into one file to calculate PSI values for local AS events.

For the identification of diff AS events, I think I might use non-parametric test and BH correction after filtering out low expression of genes and filtering out events with PSI = NA

Thanks!

Best,
Elias

[EduEyras]

that sounds a good strategy. Thanks for the feedback best Eduardo

…

On Thu, 28 Mar 2019 at 13:04, renzhonglu ***@***.***> wrote: Hi Eduardo Thank you for your reply. I will merge all the events into one file to calculate PSI values for local AS events. For the identification of diff AS events, I think I might use non-parametric test and BH correction after filtering out low expression of genes and filtering out events with PSI = NA Thanks! Best, Elias — You are receiving this because you commented. Reply to this email directly, view it on GitHub <#53 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AMWVB2tIQCTHULhyAztURQNz4n8NgE2mks5vbCNBgaJpZM4cNZpx> .

-- Prof. E Eyras EMBL Group Leader The John Curtin School of Medical Research Australian National University 131 Garran Road, Acton ACT 2601, Canberra, Australia https://github.com/comprna http://scholar.google.com/citations?user=LiojlGoAAAAJ http://www.researcherid.com/rid/L-1053-2014