Voila view problems and doubts

[Miriam Martínez]

Hi biociphers team,

while checking with voila view the results from 2 analysis I have done with majiq and moccasin, I have some doubts that I was wondering if you could help me with. I also found some problems while displaying the splicegraphs but I'm unsure if this is due to voila itself or some problem with my data.

Firstly, on the first analysis I ran, the annotation graph is not available, while on my second analysis it is (check file doubt_TANK_2). Also, on my second analysis I can't add the control group splicegraphs (check file splicegraph_problem).

Secondly, on the splicegraph of this particular event, it can be seen that it shows that this gene has 23 exons (check file doubt_TANK_4), while checking on UCSC, this gene has 8 (check file doubt_TANK_3). So, how should the data reflected on the splicegraph be interpretated? How does majiq build the splicegraphs to represent the events as sometimes show far more exons than the gene has just like this example?

Taking into account all this information, how can you know if the alternative splicing event found by the tool as significant is simply a canonical alternative splicing of an isoform or is an actual event?

Sorry if any of these questions are to basic and thank you in advance.

[San Jewell]

Hi Miriam,

First, to your questions about the splice graph differences / loading problems with the software: Along with version 2.5 of majiq, there was an ability to generate the annotation transcript above the splicegraph. I believe this was approximately a year ago. A slight while earlier than that, possibly around version 2.4, I remember there was some bug with the experiments not showing up as you displayed. I think this was a bit different per analysis type and the fix was distributed over a few versions. All that is to say: 1) it is not generally advisable to use versions of majiq/voila built/quantified with one version and viewed with another, as it may cause some undefined behavior. 2) some of these issues may be resolved using the latest version of the software. If you confirm that everything was run on the latest version and you are still seeing these missing experiments / splice graph issues, I would like to look at your experiments file and build / quantify commands, to see if I can reproduce this bug.

Now, for the latter questions.

 1) The exon numbers on the splicegraph itself are unrelated to the UCSC/canonical transcript standard. UCSC chooses a specific transcript to use as canonical which will be listed in the dropdown of annotated transcripts above the splicegraph (assuming it was in the annotation you used, probably true). These numbers will match the UCSC exon numbers. For example, in your UCSC screenshot it shows 8 exons in the transcript "ENST00000402568". If you scroll down to this transcript in the list of transcripts in voila, you will see where they align with the majiq exons. The majiq splicegraph is a collapse of all annotated transcripts provided PLUS any denovo junctions / exons / introns / exon extensions, so it will generally have equal to or greater than the number of exons in UCSC's chosen transcript. A new feature I can consider might be marking a specific transcript as automatically chosen / primary if it is marked as such in some way in your annotation file. I'm not sure exactly how UCSC chooses theirs, if it's MANE or highest TPM or something else.

2) By "actual event" in your second question, I assume you mean "novel"/"denovo". Basically we assume that the annotation you provide contains what you consider to be "canonical alternative splicing", and these appear as "annotated junctions/introns". In voila this manifests as a maroon-ish color (if there are some reads at all in the experiments) or a grey color (if there are no reads) Denovo/novel junctions, introns, exons, and exon extensions appear with a green color. "Events" or "LSVs" as we define them, may consist of completely annotated, completely denovo, or a combination of annotated and denovo connections.

Let me know if some of this is understandable.

Thanks,

-San

[Miriam Martínez]

Hi San,

sorry for the delay in answering and thank you for your clarifications, they are really helpful!

Regarding the software issues, as you point out, one of the analysis was performed with v2.4 and the most recent with v2.5, so that may explain the annotation issue. On the other hand, I still see these missing experiments / splice graph issues on both analysis and is the same issue: I can't add the splicegraphs of one of the groups (in both cases they are set as grp2 on the majiq het analysis if I recall correctly). Please let me know if I should send you the experiments file and build / quantify commands through here or separately.

Moving to the questions:

1) Thanks for the clarification on how voila represents the splicegraph, I understand better how it works now.

2) Just to check if I have understood correctly, it is possible that an "event" or "LSV" that majiq sets as significant (with the thresholds set), when using het module, can be just a difference in inclusion of that cannonical "event" or "LSV" (as it is annotated on the gff) between the groups compared. Is this correct?

Thanks!

Miriam

[San Jewell]

Hi Miriam,

Yes, for the config file / commands you use to produce the splice graph bug, if they do not contain sensitive information, you can list them here, if they do, you can email them to me directly and I will follow up here with replies.

For your "2)" question, hes, it's possible for either annotated or denovo events to pass significance thresholds, but you can also filter by the "denovo" colum if you are looking for only denovo events.

Let me know if it makes sense.

Thanks,

-San

[Miriam Martínez]

Hi San,

thank you very much for your responses, they have been very useful! Also you can find attached the config file with the experiments defined and here are the commands I used for the build and quantification (i also corrected with moccasin, if you need that info too, let me know!):

majiq build /lustre/scratch/mmartinez/replicacion_splicing_con_madmanic/genome_hg38/Homo_sapiens.GRCh38.112.gff3.gz -o /lustre/scratch/mmartinez/replicacion_splicing_con_madmanic/majiq_build/ -c /lustre/scratch/mmartinez/replicacion_splicing_con_madmanic/majiq_build_replicacion_clean_config.ini --min-experiments 0.90 --simplify 0.05 -j 4

majiq heterogen -o /lustre/scratch/mmartinez/replicacion_splicing_con_madmanic/heterogen --min-experiments 0.90 -j 4 -grp1 ${bd_file} -grp2 ${control_file} -n BD Control --stats WILCOXON

[San Jewell]

Hi Miriam,

I tried to reproduce the splicegraph selection issue using a close proxy for your configuration without using the same bamfiles and I wasn't able to yet. As such, as suggested, I'd also like to try to run the moccasin correction that you've done to rule out the edge case that this software might be causing it. Can you let me know the moccasin command that you have used?

Thanks!

[Miriam Martínez]

Hi San,

sorry for the late response! It seems that I don't have the notifications on and I just came across with your answer while checking the google group. Here is the model matrix (attached) and the command I used for moccasin correction (I also include the python version in case it is important):

module load python/3.10.13

python moccasin.py /lustre/scratch/mmartinez/replicacion_splicing_con_madmanic/model_matrix_replicacion_clean_gender_only.txt /lustre/scratch/mmartinez/replicacion_splicing_con_madmanic/majiq_build /lustre/scratch/mmartinez/replicacion_splicing_con_madmanic/corrected_sj_files gender

In addition, maybe this information is also useful for debugging: while rechecking the results with voila, I could check that if I kept scrolling down the list, in the end, appeared the file names and could add them to the splicegraph.

If you need anything else, please let me know!

Best,

Miriam