Voila view Combined track is not an addition of replicate samples

[Steven Baker]

Hey y'all,

I'm trying to understand where the numbers across splice junctions come from in Voila view output. 

I ran a dPSI analysis in MAJIQ 2.5 with 2 experimental groups (WT or DUB), each containing 2 biological replicates. When visualizing with Voila view, I will typically compare WT combined to DUB combined. I thought that the numbers above splice curves were an indication of junction spanning reads from short read RNA-seq, and thus thought that the combined track would simply add up WT replicate 1 and WT replicate 2. However, this is not the case. Sometimes it looks like the combined track is simply reporting the numbers from 1 replicate, whereas in other scenarios they are unique numbers that don't really make sense. Please shed some light for me.

A few images attached. Many thanks for your help and cultivating this community.

Steve

[San Jewell]

Hi Steve,

Please note that there may be more experiments available to add as additional splicegraphs from the "splice graph menu" , I'm not sure we are seeing them all in this screenshot. I believe the decision was made in moving from majiq 2.0 to 2.3 or 2.4, that instead of the sum across experiments per group, we instead show the median across experiments per group. Let me know if this makes more sense after looking at all of the splice graphs available in the build. 

This change was proposed by a few members of the lab as being of more utility, however, if there is significance to seeing the sum instead of the median as an option, please let me know and we can potentially add this as an option in future versions.

Thanks,

-San

[Steven Baker]

Thanks San. Median makes sense and FWIW I do think an *option* for sum would be useful, though I agree with your team that median is more useful. Indeed as you point out there are other tracks to display, but for simplicity, I only elected to show the 3 tracks from WT (combined, rep1 & rep2 [here titled 3 & 4, respectively]) and not the 3 tracks from DUB (combined, rep1, rep2).

However, using median does not seem to reconcile what I'm seeing in my splicegraphs. 

In case the resolution isn't good enough from the bottom image, I'll tell you the splice numbers. Combined, 5'internalSS (3), 5'terminalSS (11); rep1, 5'internalSS (0), 5'terminalSS (2); rep2, 5'internalSS (3), 5'terminalSS (0). Firstly, this to me shows that the variance is so great between the 2 experimental replicates that this should not have been called on dPSI. Secondly, the combined values are certainly not the median of these two experimental replicates. Am I missing something?

Cheers,

Steve

[San Jewell]

Hi Steve,

For the bottom image it doesn't look like medians are represented for *any* of the splice junctions in the image, so I think it is the case that there are more splice graphs / experiments in this build. Could you check in the splice graph menu if there are more experiments to add? If it still doesn't make sense can you add more screenshots with all of the splice graphs added? (Or multiple images to split it up if there are too many)

Thanks,

-San

[Steven Baker]

Here's a few screenshots showing that all files are loaded. Below are the contents of my .ini for majiq build

Appreciate your help!

Steve

[info]
bamdirs=/Users/stevenfbaker/Docs/Bioinformatics/Whitehurst/Source/human
genome=hg38

[experiments]
WT=T1-WT-3,T1-WT-4
DUB=T1-DUB-3,T1-DUB-4

[San Jewell]

Hi Steve,

Thank you for the comprehensive information, after experimenting a little bit I was indeed able to reproduce this issue, and I have pushed out an update to the repository to address it. Please install or update using the same method and try to run "voila view" again. Note that the issue was just in the webfrontend part, so there is no need to re run any build / dpsi analysis. Please let me know if you have tested and found the issue resolved.

Thank you for catching this.

-San

[Steven Baker]

Thanks San. Do I just need to update Voila or does Majiq also need updating? Do you know the commands for updating?

[San Jewell]

Hi Steve,

The command for pip install would work the same to update as to install after activating your python environment. Technically only voila needs to be updated, not majiq. I avoid posting the raw command here because that may allow some to avoid agreeing to the license terms, however, there is no issue with using the same email on the https://majiq.biociphers.org/app_download/ page to regenerate the command / install page. Please let me know if it makes sense.

Thanks!

-San

[Steven Baker]

Thanks for the fix, San. I was able to reinstall and voila view looks better now. 

It seems that this is certainly not the most efficient way to troll through my data. Are there any tutorials for data analysis from majiq/voila outputs?

Cheers,

Steve

[San Jewell]

Hi Steve,

I would recommend probably making a new topic for this is possible.

Can you describe your study case or what kind of filters you are looking for? There are a number of modes and it's difficult to give advice without any context. For example there is a sample use case in the "gallery" section of the documentation which runs through many of the usage modes for a heterogen-based study. (https://biociphers.bitbucket.io/majiq-docs/gallery/heterogen-vignette.html)

Let me know,

-San