Too few reads in bam for Sniffles about long-read-pipelines HOT 10 OPEN

Hi Brian, thanks for the report.

Do you know the coverage of this input bam?

I have a suspicion that this bam might have a low coverage, hence Sniffles cannot estimate error rate and other parameters from the few reads.

from long-read-pipelines.

@SHuang-Broad I think you're right.

I wasn't able to find any specific coverage report for this bam, so I ran samtools idxstats on it and it appears it has one read aligned on chr2, and that's it. So it's not terribly surprising it failed.

I guess it would be nice if the pipeline failed a little more gracefully and maybe was able to continue on? But there's really nothing here to do much with...

from long-read-pipelines.

Wow, one read? Is this a test BAM file or real data?

from long-read-pipelines.

@brigranger hmm.... Typically variants is the last step for the pipelines now, so if this one particular task fails, no downstream tasks are severely impacted practically speaking.

But I do think there could be more QC done at the beginning (or maybe middle) of the workflow, i.e. quit early when things are really suspicious.

What do you think, @kvg ?

from long-read-pipelines.

This is an interesting one. Thinking ahead to amplicon sequencing, there will certainly be cases where we parallelize over chromosomes and some will have no data. So it looks like we'll have to figure out a way to protect ourselves from those kind of failures when the tools we're running don't have those protections built-in themselves.

How to do that robustly will require some thought. @brigranger can you send us a link to the original input BAM?

from long-read-pipelines.

@kvg You should be able to find it here: gs://broad-gp-pacbio/r64020_20200116_203442/2_B01/

from long-read-pipelines.

I looked at the subreads bam, it seems to be a low yield CCS flowcell.

gsutil du -sh gs://broad-gp-pacbio/r64020_20200116_203442/2_B01/m64020_200118_025318.subreads.bam
151.93 GiB   gs://broad-gp-pacbio/r64020_20200116_203442/2_B01/m64020_200118_025318.subreads.bam

So it looks like something in CCS that produced ultralow depth.

from long-read-pipelines.

@kvg , yes, I believe this is ultimately tied to what QC we put in each step of the pipeline, as by nature has to be an implement-as-we-encounter issue, considering that there could be many places where things go wrong.

from long-read-pipelines.

@brigranger is this the only occasion where you see the error?
If so, I'm going to close this and deal with #113 instead as a solution.

from long-read-pipelines.

from long-read-pipelines.