Detection problem of tandem cassette exons by voila modulizer

[Ryo Kurosawa]

Hi, MAJIQ developers,

I try to detect tandem cassette exon events by majiq deltapsi and voila modulizer.

I found one three-exon skipping was apparent from the result of majiq deltapsi.

However, that was not detected as a tandem cassette exon by voila modulizer although each lsv has very high delta psi and significant probabilities.

Only another two-exon skipping was detected in the same lsv area.

I would like to see the added Power point and Excel files for details.

Q1. Why the three-exon skipping was not detected??

Q2. Why does tandem_cassette.tsv have some blank values in some columns (lsv_id, junction_changing, any psi values for example). Are not they significant tandem cassette exons?? (The detected two-exon skipping above also has some blanks)

This is the command I executed.

voila modulize  \
splicegraph.sql deltapsi.voila \
--overwrite \
-d ./voila \
--changing-between-group-dpsi 0.05 \
--non-changing-between-group-dpsi 0.005 \
--probability-changing-threshold 0.9 \
--keep-constitutive \
-j 8

Regards,

Ryo Kurosawa

[San Jewell]

Hello Ryo,

As a first debugging step I would recommend removing some of the filters to make sure you get what you expect when we are not removing possibilities based on quantifications, and then we can go from there to see which filter is removing your junction(s) of interest.

Could you please add the switches to your command --keep-no-lsvs-junctions --show-all

And check if you can see the structure then? The missing values usually indicate missing quantifications.

-San

[Ryo Kurosawa]

Hello San,

Thank you for your suggestion.

However, even with these options, the detected structure in FKTN gene was not changed at all.

Of course, in the other region, some events were newly detected.

(About 4331→5189 event_ids with these options)

Do you have any other ideas to resolve this problem?

Regards,

Ryo Kurosawa

2023年4月21日金曜日 0:32:03 UTC+9 sje...@biociphers.org:

[Matthew Gazzara]

Hi Ryo,

Thanks for your questions and trying out the Modulizer. 

In your original file for the TCE event involving exons 15 through 18, the missing LSV IDs in those lines means that that LSV was not detected or quantified in your input. The Modulizer can still use the junction information from the splicegraph even if there may be missing LSV quantifications. I suspect the target LSV between exon 18 and 15/17 is not quantified in your delta psi input because of the low number of junction spanning reads in the second sample of you splicegraph (two reads from e18 to e15 and either four or one reads from e18 to e17). By default we require a sum of I believe 10 reads summed across the junctions of an LSV for it to be quantifiable in a condition. Your second condition does not meet that criteria and so you don't see it in the output when comparing condition 1 versus condition 2. 

The exon 16 through exon 18 TCE you predict looks like it should be detected in some form. Because there are a lot of other splice junctions involved in this region of the splicegraph we may be putting it into a more complex event type (multiexon skipping) which sort of acts like a catch-all for these complex things. TCE is a bit more rigid of a definition where we don't allow additional connections or event types happening between or around the tandem cassettes...although it is a bit curious that we detect the two exon version you show in your example as a TCE, but not the three exon version. If it was being shuttled into the multi exon skipping file I would expect both versions to be there.  

Can you share all the deltapsi.tsv lines for FKTN for me to look through for other LSV quantifications? Also I think it would be helpful if you shared the FKTN lines from the Modulizer summary.tsv file from your original run and the second run form San's suggestions. This could help us figure out what type of event (if any) the three exon TCE event you are interested in is being categorized as.

Thanks,

Matt 

[Ryo Kurosawa]

Hi Matt,

Thank you for the more detailed discussion.

I added deltapsi.tsv, summary.tsv(original/secondrun), and tandem_cassette.txt(original/secondrun) to this message.

The second run appears only to generate duplicated event ids.... 

>By default we require a sum of I believe 10 reads summed across the junctions of an LSV for it to be quantifiable in a condition.

Is this restriction removed on "--keep-no-lsvs-junctions" and "--show-all" options??

>TCE is a bit more rigid of a definition where we don't allow additional connections or event types happening between or around the tandem cassettes

Thank you for telling me the strict definition. I wanted to cover all the splice events expected from junction reads, and recognize that MAJIQ is the only possible program to realize it !!!

Regard,

Ryo Kurosawa

2023年4月22日土曜日 5:05:28 UTC+9 mgaz...@biociphers.org:

[Ryo Kurosawa]

Hello.

Was the potential bug fixed about tandem cassette exons ??

Regard,

Ryo Kurosawa 

2023年4月24日月曜日 18:09:13 UTC+9 Ryo Kurosawa: