genotype_sv crash with v2.4

[marqueda]

I tried running genotype_sv with the new release 2.4, but it immediately crashes and issues the following error message "cannot create std::vector larger than max_size()". I tried the same command with the older release 2.2 and it started running fine.

Here was the command I used (I renamed the binayr graphtyper2.4 on my machine):

graphtyper2.4 genotype_sv ref.fasta manta.SVs.vcf.gz --sams=bams --region_file=region.9.file --threads=1 --log GraphTyperSV.Run.9.log -v --output=allGenomes.SV

Given that I used all bam files, it would be not that easy to share the data... let me know if I need to try it again on a smaller dataset able to share with you for reproducing the issue.

Best, David

[marqueda]

Just tried the release version v2.3 and get the same segfault and message.

[hannespetur]

Hey David, that's odd. I can't find this error message anywhere in the code. Can you run with --vverbose (very verbose) and --log=log and send me the log file so I can get a better idea where it might be occuring.

Best, Hannes

[marqueda]

Thanks for the super fast reply! Here is the logfile.

[hannespetur]

Thank you. Looks like its in the graph construction, which is good since then I won't need any bams to reproduce. Are you able to share the VCF? Only the region chr9:1-1200000 should be enough.

Best, Hannes

[marqueda]

Sure, that's a quite small file. This file includes svimmer merged Manta calls and I applied the awk-fix from an earlier issue.

Same result though with a VCF file without the awk-fix. I also tried another version where I merged SNPs / short SVs from graphtyper genotype with the Manta / svimmer merged large SV calls. I merged them with bcftools concat -a -d all and sorted them with bcftools sort. Maybe that would give additional problems in case of duplicate positions / overlapping positions?

[hannespetur]

Sorry, I forgot that I also need the reference FASTA. Could you upload that as well?

Best, Hannes

[marqueda]

Sure, here it is: Link. Please let me know once you have downloaded it - I will take it down again, as it is still unpublished...

[hannespetur]

Thank you. Download is finished.

[hannespetur]

Seems like I was creating a vector with some invalid iterators in some very rare cases 😕 Looks like its fixed now. I guess the error message is somewhere within the c++ stl.

Could you try this very unofficial graphtyper binary? I would like to make some more tests before making a proper release but I hope this will do in the meantime.

Best, Hannes

[marqueda]

Great, thanks fixing this bug! It is working now, at least not an immediate segfault anymore. I will let you know whether it finished calling the whole chromosome.

[marqueda]

Dear Hannes, Sorry for the slow reply, our cluster had an issue and I wasn't able to run jobs until yesterday. I am now getting another crash with the beta version you posted above, with the error message being "std::bad_alloc".

I am running the following code:

./graphtyper_beta genotype_sv ${ref} ${vcf} --sams=bams --region_file=region.${LSB_JOBINDEX}.file \
--threads=34 --log GraphTyperSV.Run.${LSB_JOBINDEX}.log -v --output=allGenomes.SV --vverbose

And I have provided 34 cores that allow hyperthreading and 85 GB memory. The reporting also notes that the job took max. 56 GB memory and ca. 70% of the cores I reserved in the submission system. Do you have an idea where the allocation error might arise from? Here is the --vverbose logfile output by beta version graphtyper: Link.

Thank you for your help! David

[hannespetur]

Hey David. It seems there was another bug in the version I posted previous and I think it might be the cause. But I have now fixed it. Can you try the new version 2.5?

https://github.com/DecodeGenetics/graphtyper/releases/download/v2.5/graphtyper

Thank you very much for the all the feedback, it has been really helping me spotting my mistakes.

Best, Hannes

[marqueda]

Dear Hannes, You are welcome and thank you back for the quick responses and bugfixes! I have tried running the new release, but it crashes with the error message "vector::_M_range_check: __n (which is 5141) >= this->size() (which is 5141)". Here is the --vverbose log file again: Link. Best, David

[hannespetur]

Ok, let's give this version a shot at it: https://www.dropbox.com/s/5im1ta8t77b3ck2/graphtyper?dl=0

[marqueda]

Thanks! Unfortunately, our cluster login nodes were just shut down due to an alleged cyper-attack on many HPC systems worldwide... I will let you know once I can access everything again and run the next test.

[hannespetur]

Sorry to hear that, I hope the situation will be resolved soon.

[marqueda]

Back in the game. The beta version runs (no segfault) but issues the following error message shortly after starting, after which graphtyper shuts down:

[E::hts_idx_push] Unsorted positions on sequence #1: 8840 followed by 7567
cp: cannot stat ‘/scratch/123193665_9.tmpdir/graphtyper_200515_081919_chr9_000000001.6yLLm3/graphtyper.vcf.gz.tbi’: No such file or directory

The last two logfile lines are the following, full log file here again: Link.

[2020-05-15 08:12:33.027060] <warning> vcf.cpp:1191 Could not build VCF index
[2020-05-15 08:12:36.216295] <error> This command failed 'cp -p /scratch/123192880_9.tmpdir/graphtyper_200515_080150_chr9_000000001.2YIUaV/graphtyper.vcf.gz.tbi allGenomes.SV/chr9/000000001-001000000.vcf.gz.tbi'

I tried to load the samtools / tabix module in the job on our cluster, but the same error message appeared, so I assume it cannot find the dependency?

I also checked the input VCF file with structural variants and position 7567 is actually before 8840, and I have sorted it before with bcftools sort. So I don't think this could be the issue?

[hannespetur]

It won't need to find the dependency, htslib (which has tabix functions) is statically linked with graphtyper so the code is there. It must be something else. Could you send me /scratch/123192880_9.tmpdir/graphtyper_200515_080150_chr9_000000001.2YIUaV/graphtyper.vcf.gz ? (it is enough to only give me the variant site information which is the first 8 columns of the VCF).

[hannespetur]

Nevermind, no need to send me the VCF. I simulated some reads on your reference and could reproduce your problem. It should now be fixed in the new v2.5.1.

Best, Hannes

[marqueda]

Dear Hannes,

Sorry for the late reply, I was offline since Friday mid-day. Unfortunately our computer cluster has been closed again due to the cyber attack on ~16 European HPC centers. It's unknown when they will open again (days to weeks), but I will let you know once I gave the new version v2.5.1 a try!

Thanks again and best, David

[marqueda]

Dear Hannes, Our cluster is back to life and graphtyper_sv 2.5.1 is running successfully, both on SVs only and on a VCF file with SNPs/indels and Manta SVs. Thanks for fixing the issue! Best, David

[hannespetur]

Great! Thanks.