the two junctions in some 2-junction LSV only has 1bp difference

[Maoting Chen]

Hi San,

I recently took a closer look at the splicing results from MAJIQ v3 and noticed that there are quite a few 2-junction LSV where the involved two junction only have 1bp difference. For example, the LSV gene:WBGene00003887:s:13578172-13578283 contains two junctions: 13578283-13578344;13578284-13578343. So both the junction starts and ends of the two junction only have 1bp difference and they don't share the same donor/acceptor sites. Such LSV is not detected in MAJIQ v2 results using the same bam files for analysis.

Are those LSVs real? How to interpret LSVs like this?

Thank you,

Maoting

[San Jewell]

Hi Maoting, 

It's difficult to get a picture of the splicing that is going on from only the information provided. Do you have a voila view of the splicegraph? One thing I would think is likely, is that you are seeing more retained intron events (intron coordinates are one-off from exon coordinates). The reason you may see more in v3 is due to changes in how annotated introns are preserved. Are the events that you see intron events? 

Thanks, 

-San

[Maoting Chen]

Hi San,

Thank you for your reply.

I am using University cluster to run MAJIQ, but I haven't figured out how to open the html from voila view running on a remote computer node. 

You are right that those LSVs are indeed intron retention events. For the example I provided, the junction coords are 13578283-13578344;13578284-13578343, and 13578283-13578344 is the junction across the annotated intron and 13578284-13578343 is within this annotated intron. So the difference caused in MAJIQ v3 from MAJIQ v2 is from the change how majiq build works? Could you please explain a bit more why the change happens and how to justify if this captures real biological change? And if there is flexibility to disable the change?

Thank you,

Maoting

[Maoting Chen]

Hi San,

I was able to get the voila view running on the cluster. Please see the attached figures for two examples. Indeed, those are intron retention events. Looking forward to your response to my previous questions.

Thank youm

Maoting

[Maoting Chen]

Additional question is that: I ran voila view on one group of condition and it contains 4 replicates. However, in the html only the read count shown seems to be only from one replicate OP1. As you see the below example, the read count shown for the LSV is so few and PSI is no quantified. But from MAJIQ output on the whole group, the LSV has PSI quantified. Is it because I am using a lower version of voila -- Voila v3.0.18.dev1+g69f22130d? 

Thank you,

Maoting

[San Jewell]

Hi Maoting, 

I appreciate your continued feedback on majiq and voila, and I hope I can answer your questions satisfactorily. 

-In majiq v2, there was a prior given to annotated junctions, allowing their LSVs to be quantified as if there was a splicing event at that point (annotation is given tiny prior in the calculation), you can see this behavior for 0 read annotated junctions in both v2 and v3. In v2, this same behavior was not applied to annotated introns. The difference between v2 and v3 is that now, introns have this prior in addition to junctions, which will cause many more introns to be quantified with very low PSI in the output.

-For question 2, I see and note the problem with voila. I have determined there is an issue with psi vs multipsi modes when using certain combination of one psi file with many prefixes inside it vs many psi files with one or more prefixes. I am going to add more explicit usage cases/functionality to clear this up as a high priority item. However, in the meantime there is a way to get a functionality similar to v2 (means over multiple experiments):

run $ majiq-v3 psi-coverage ; once for each of your replicates, this will provide four .psicoverage files ; by default, you can run voila with these four files, but it will show their psi separately (may still be interesting to you) ; however, you can show the grouped values by specifying a grouping file to voila (defines how to group psicoverage files together in a voila view); in this case the file would just have a single group defined. The contents of the file is a simple .tsv with columns 'group' and 'prefix' ; for example, using the workshop example data, for the three Adr samples the file would look like the one attached. You specify the file to voila using the --psicov-grouping-file switch. Note: I am aware this is a workaround and more cumbersome than v2, so the default functionality will be improved shortly and made more clear in the documentation. Using voila this way will also show median reads as well as optional splicegraphs for each experiment. 

Please let me know if this method helps for your case. 

Thanks!

-San

[Maoting Chen]

Hi San,

Thank you for your detailed clarification.

It has been really helpful for me to understand how MAJIQ works and how to use MAJIQ better.
I'll keep an eye out for the updated of majiq/voila.

Thank you,

Maoting