Cuffdiff 2.2.0 segmentation fault

[blancha]

Hi,

I had the bad idea of switching to Cufflinks 2.2.0 and the new format with a sample sheet and a contrasts file. I'm now getting segmentation faults in all my analyses at the testing for differential expression stage. Any clues as to what is wrong with my command, which I've posted below, or if it might be a problem with the program or the cluster?

***

tail cuffdiff.sh_error

> Processing Locus 1:3102015-3102125 [ ] 0%/var/spool/pbs/mom_priv/jobs/15151606.gm-1r16-n04.guillimin.clumeq.ca.SC: line 36: 23285 Segmentation fault (core dumped) cuffdiff --use-sample-sheet -C contrasts.txt -p 12 --library-type fr-firststrand -u -b /sb/project/afb-431/genomes/Mus_musculus/Ensembl/GRCm38/Sequence/WholeGenomeFasta/genome.fa -o ../../cuffdiff /sb/project/afb-431/genomes/Mus_musculus/Ensembl/GRCm38/Annotation/Genes/Mus_musculus.GRCm38.75.gtf sample_sheet.txt

***

cuffdiff.sh

#PBS -l nodes=1pn=12
#PBS -l walltime=4:00:00:00
#PBS -o ${PBS_JOBNAME}_output
#PBS -e ${PBS_JOBNAME}_error
#PBS -V
#PBS -N cuffdiff.sh
#PBS -q sw
#PBS -m bae

cd $PBS_O_WORKDIR

export OMP_NUM_THREADS=12

cuffdiff \
--use-sample-sheet \
-C contrasts.txt \
-p 12 \
--library-type fr-firststrand \
-u -b /sb/project/afb-431/genomes/Mus_musculus/Ensembl/GRCm38/Sequence/WholeGenomeFasta/genome.fa \
-o ../../cuffdiff \
/sb/project/afb-431/genomes/Mus_musculus/Ensembl/GRCm38/Annotation/Genes/Mus_musculus.GRCm38.75.gtf \
sample_sheet.txt

***

sample_sheet.txt

sample_id group_label
../../tophat/DNAM_NE/accepted_hits.bam DNAM_NE
../../tophat/DNAM_PO/accepted_hits.bam DNAM_PO

***

contrasts.txt

condition_A condition_B
DNAM_NE DNAM_PO

***

cuffdiff.sh_output

----------------------------------------
Begin PBS Prologue Sun Apr 13 08:47:14 EDT 2014 1397393234
Job ID: 15151606.gm-1r16-n04.guillimin.clumeq.ca
Username: blancha
Group: afb-431-01
Nodes: sw-2r15-n32
End PBS Prologue Sun Apr 13 08:47:14 EDT 2014 1397393234
----------------------------------------
----------------------------------------
Begin PBS Epilogue Sun Apr 13 11:12:36 EDT 2014 1397401956
Job ID: 15151606.gm-1r16-n04.guillimin.clumeq.ca
Username: blancha
Group: afb-431-01
Job Name: cuffdiff.sh
Session: 23131
Limits: neednodes=1pn=12,nodes=1pn=12,pmem=2700mb,walltime=96:00:00
Resources: cput=05:11:19,mem=9762812kb,vmem=10809000kb,walltime=02:25:22
Queue: sw-parallel
Account: afb-431-ac
Exit code: 139
Nodes: sw-2r15-n32
Killing leftovers...

End PBS Epilogue Sun Apr 13 11:12:37 EDT 2014 1397401957

***

Thank you for your help,

Alexis

[yueluo]

Try use full-path for all your input. And where does $PBS_O_WORKDIR point to ?
If the sample sheet and contrast files don't work, you can always go back to the old command line :
cuffdiff [options]* <transcripts.gtf> sample1_replicate1.sam[,...,sample1_replicateM.sam]>
Since you only have two bam files as input, then sample sheet and contrast file seem unnecessary.

[blancha]

Thanks for your reply.

Yes, I've just finished running Cuffdiff using the old command line format.
It worked fine.
I do however, have other analyses with a larger amount of conditions and experiments where the new format would be useful, and I also get a segmentation fault at the testing for differential expression stage.
I only posted this example because it is simpler.

$PBS_O_DIR is a PBS variable that refers to the directory from which the queue submission is done. It does seem to refer to the correct directory, since that is where the error file is put.

The paths to the files seem to be correct, since the first stage, the abundance estimate works. I'll still try with the full paths, to see if that helps.

[05:49:31] Calculating preliminary abundance estimates
> Processing Locus 1:3773817-3773879 [ ] 0%
> Processing Locus 1:4481008-4486494 [ ] 0%
> Processing Locus 1:4797868-4887987 [ ] 0%
...
> Processing Locus 1:168875727-168876489 [* ] 4%
...

I've posted below the same error message at the differential expression stage for another analysis.

***

[11:59:57] Testing for differential expression and regulation in locus.
> Processing Locus 1:3092096-3092206 [ ] 0%/
var/spool/pbs/mom_priv/jobs/15140816.gm-1r16-n04.guillimin.clumeq.ca.SC: line 36
: 28419 Segmentation fault (core dumped) cuffdiff --use-sample-sheet -C con
trasts.txt -p 12 --library-type fr-firststrand -u -b /sb/project/afb-431/genomes
/Mus_musculus/Ensembl/NCBIM37/Sequence/WholeGenomeFasta/genome.fa -o ../../cuffd
iff /sb/project/afb-431/genomes/Mus_musculus/Ensembl/NCBIM37/Annotation/Genes/ge
nes.gtf sample_sheet.txt

***

[yueluo]

I have experienced the same error when I first ran cuffdiff from cufflinks-2.2.0.
So far I haven't figured out a way to 'safely' run the new workflow(including cuffquant and cuffnorm)... I also ran into some other problems, which I started a post about with no replies so far. I guess this version is still kind of 'beta'.

[mattia]

I have the same kind of error: I've tried to launch cufflink 2.2.0 with the same parameters of previous versions (using a bam file already analyzed!!!), and a "segmentation fault" error occurred.

...
...
[11:19:51] Learning bias parameters.

> Processing Locus chr1:11868-14409            [                         ]   0%/var/spool/PBS/mom_priv/jobs/10652.srv-hngen.SC: line 21: 12458 Segmentation fault      cufflinks -p 8 -u -b $hg_ref --compatible-hits-norm -G $annotation_trans -o $aaq/$ssq/ --library-type fr-secondstrand $ccq/$ssq/accepted_hits.bam

Hope this error will be fixed quickly!!!!!

Mattia.

[blancha]

Well, I've isolated the problem at least. I can run Cuffquant without a problem. I can run Cufflinks with the sample sheet file but without the contrasts file. The problem seems to be with the contrasts file and the contrasts parameter.

I get a segmentation fault when I specify the contrasts file. If I remove the contrasts parameter, and let Cuffdiff do all possible pairwise comparisons, the run completes.

I've found at least one other user reporting the same problem.

https://groups.google.com/forum/#!msg/tuxedo-tools-users/-lCUNFJe_2c/HhTtYjLmhs0J

Even without the contrasts file, this new workflow is a lot faster, since I can now further parallelize the counting by using one Cuffquant instant per sample, and the runtime of Cuffdiff is a lot shorter.

I hope this error gets fixed in the next update. I sometimes feel it would be quicker to rewrite the source code for the Cufflinks suite rather then spending so much time learning all its quirks.

[yueluo]

Good to hear blancha. I posted this problem on April 1st and no one ever replied. Guess I picked the wrong day to post :-(

Anyway, I'll try re-running cuffnorm with 'cuffdiff' output format and see what happens. It was really frustrating to see 4 runs with 3 'kinds' of tables(2 of which where re-ordered).

[1520191]

cuffmerge -g Mus_musculus.GRCm38.75.gtf -s musref.fa -o merged_asm -p 8 assemblies.txt

[Mon Apr 28 17:08:50 2014] Beginning transcriptome assembly merge
-------------------------------------------

[Mon Apr 28 17:08:50 2014] Preparing output location merged_asm/
[Mon Apr 28 17:08:52 2014] Converting GTF files to SAM
[17:08:52] Loading reference annotation.
[17:08:52] Loading reference annotation.
[Mon Apr 28 17:08:52 2014] Quantitating transcripts
You are using Cufflinks v2.2.0, which is the most recent release.
Command line:
cufflinks -o merged_asm/ -F 0.05 -g Mus_musculus.GRCm38.75.gtf -q --overhang-tolerance 200 --library-type=transfrags -A 0.0 --min-frags-per-transfrag 0 --no-5-extend -p 8 merged_asm/tmp/mergeSam_file22R91w
[bam_header_read] EOF marker is absent. The input is probably truncated.
[bam_header_read] invalid BAM binary header (this is not a BAM file).
File merged_asm/tmp/mergeSam_file22R91w doesn't appear to be a valid BAM file, trying SAM...
[17:08:58] Loading reference annotation.
[17:09:06] Inspecting reads and determining fragment length distribution.
Processed 29389 loci.
> Map Properties:
> Normalized Map Mass: 29421.00
> Raw Map Mass: 29421.00
> Fragment Length Distribution: Truncated Gaussian (default)
> Default Mean: 200
> Default Std Dev: 80
[17:09:06] Assembling transcripts and estimating abundances.

8:119910359-124345724 Warning: Skipping large bundle.
Processed 29388 loci.
[Mon Apr 28 17:09:23 2014] Comparing against reference file Mus_musculus.GRCm38.75.gtf
You are using Cufflinks v2.2.0, which is the most recent release.
GFF Error: duplicate/invalid 'transcript' feature ID=ENSMUST00000105372
[FAILED]
Error: could not execute cuffcompare

Hi, I have this kind of error,i dont know how to fix this problem? can someone recommend ed a good ideal to solve it? thanks a lot .