Bug in multiqc / mirTop broke

[apeltzer]

Had this unfortunately with the dev branch:

  /// MultiQC <U+1F50D> | v1.12

|           multiqc | Search path : /work/38/b3bd004777869168b75aaf74ee8d69
|         searching | ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% 418/418
|    custom_content | software_versions: Found 1 sample (html)
|    custom_content | nf-core-smrnaseq-summary: Found 1 sample (html)
╭───────────────── Oops! The 'mirtop' MultiQC module broke... ─────────────────╮
│ Please copy this log and report it at                                        │
│ https://github.com/ewels/MultiQC/issues                                      │
│ Please attach a file that triggers the error. The last file found was:       │
│ ./full_mirtop_stats.log                                                      │
│                                                                              │
│ Traceback (most recent call last):                                           │
│   File "/usr/local/lib/python3.10/site-packages/multiqc/multiqc.py", line 65 │
│     output = mod()                                                           │
│   File "/usr/local/lib/python3.10/site-packages/multiqc/modules/mirtop/mirto │
│     self.parse_mirtop_report(f)                                              │
│   File "/usr/local/lib/python3.10/site-packages/multiqc/modules/mirtop/mirto │
│     parsed_data["read_count"] = parsed_data["isomiR_sum"] + parsed_data["ref │
│ KeyError: 'ref_miRNA_sum'                                                    │
│                                                                              │
╰──────────────────────────────────────────────────────────────────────────────╯
|          samtools | Found 158 stats reports
|            fastqc | Found 79 reports
|           multiqc | Compressing plot data
|           multiqc | Report      : multiqc_report.html
|           multiqc | Data        : multiqc_data
|           multiqc | Plots       : multiqc_plots
|           multiqc | MultiQC complete
|           multiqc | 1 flat-image plot used in the report due to large sample numbers
|           multiqc | To force interactive plots, use the '--interactive' flag.
See the documentation.

Not sure what is causing this, didN#t check the module so far.

@ewels any idea?

[ewels]

Not sure sorry, I guess the output is different for some reason? Not seen it before, need to see the tool output. These things are usually fairly easy to patch though.

[apeltzer]

Hi @ewels I have added some fake data (that resembles the original data very much) to add a testcase for mirtop mabye. See linked PR to testdata: MultiQC/test-data#227

@lpantano might also be interested in seeing this -- it looks like the ref_miRNA_sum variable is not present for all the samples here, which might explain that the access then fails.

[lpantano]

oh, I didn't think about not having that for all samples...that seems weird to be honest, related to the data itself...no sequences matching the references seems weird. Thank for spotting this.

[apeltzer]

Considering that we likely need to address this upstream either in mirtop to enforce this is always there in the output or in the multiqc module - can we potentially add a --skip_mirtop option to enable users to circumvent this bug for the time until this is fixed upstream? @JoseEspinosa @lpantano what do you think about that?

[JoseEspinosa]

Yes, might be a good solution for now. Should we skip the whole process or maybe only multiqc i.e. --skip_mirtop_multiqc ?

[lpantano]

I think the quantification is useful, I would prefer to skip the QC of multiqc in this case only if it were my data.

[lpantano]

I fixed that in mirtop, waiting for bioconda to bump the version, and then we need to update the docker container, that I forgot how to do it, so If somebody knows how, happy to get help here :)

[JoseEspinosa]

You mean the mulled container here ? I can take care of it if you are referring to this

[JoseEspinosa]

Here is the PR in the biocontainers multi-package containers repository BioContainers/multi-package-containers#2150

[JoseEspinosa]

Here is the PR with the updated container #143

[apeltzer]

All fine now 👍🏻

[apeltzer]

Hmm, unfortunately still an issue. I've had more projects with the same error :-(

[apeltzer]

It still looks like some samples are missing the same entry, e.g. ref_miRNA_sum is not present for all samples unfortunately. Thus the MultiQC module breaks --> crashing pipeline :-(

[gpalidwor]

I'm getting the same problem, looks like same cause; smrnaseq pipeline failing on same mirTop parsing error, rest of the pipeline completes. Can you recommend a workaround, is there a way to suppress multiqc parsing of mirTop in the pipeline or should I just run multiqc separately?

[apeltzer]

I think the problem is, that the MultiQC module assumes certain keys to be present in the mirtop output - however mirtop does not always produce these, causing the MultiQC module to crash as it can't find these for all samples when computing metrics. We hoped that the fix @lpantano had made in mirtop 0.4.25 fixed this, but apparently not :-(

Currently not really a workaround - skipping multiqc and running it outside means you will not have a nice report which is also not too nice.

[ewels]

Is there an issue / PR already for this on the MultiQC repo? Should be fairly easy to fix.

[gpalidwor]

Is there an issue / PR already for this on the MultiQC repo? Should be fairly easy to fix.

I don't see one, want me to open it?

[ewels]

Yes please 👍🏻 Then maybe we can put @ErikDanielsson on the job over a coffee break 😅

[gpalidwor]

Bug opened MultiQC/MultiQC#1712

[apeltzer]

Maybe a good idea yeah to fix this in MultiQC, although that means we have to wait for MultiQC V1.13 to fix this 😢

[ewels]

[lpantano]

Hi,
if you give me a gff file from mirtop that is missing that I can take a look. I know where the code is so I can pull a fix quickly. It is working for us, so I don't have any example.
Thanks!

[gpalidwor]

below is the gff file from the mirtop output in the results directory of the nf-core/smrna output from the run where the multiqc parsing of the mirtop output failed. It's mostly zero counts because it was a bad configuration, rerunning to see if multiqc succeeds with the proper config

mirtop.gff.gz

[gpalidwor]

fwiw: my issue was due to a misconfiguration leading to samples with low or zero mapping. Fixing the configuration fixed the multiQC/mirtop problem for me. This is probably still a multiQC issue that should be fixed eventually as a lack of any mapped miRNA is a reasonable edge case.

[ewels]

Yeah, MultiQC should never crash with a traceback like that. For starters it means that all subsequent logs from that tool will be ignored, so it should definitely be fixed 👍🏻

[ewels]

Ok @ErikDanielsson should have fixed it in MultiQC, just merged to dev.

If you want this quickly you can do a Bioconda release of dev if you like. People have done that in the past.

[apeltzer]

I'll do that then, should be fine to do in such a case and then upgrade in smrnaseq once we have 1.13 again 🥳

[apeltzer]

PR in modules is open nf-core/modules#1814 , once this is merged, can upgrade the module for MultiQC here in the repository and we're fine again 👍🏻

[apeltzer]

MultiQC 1.13 is released, as such this should be fixable by upgrading to the MQC 1.13 module