[hughfrench]

Hi Dessi,

Come see me tomorrow,

hugh

[ug14cxb]

As someone who analyses RNAseq data I would recommend not using cuffdiff for detecting differentially expressed genes. There are several studies that show it lacks precision. Here is one example.

http://bib.oxfordjournals.org/content/early/2013/12/02/bib.bbt086.long

I would recommend eXpress for expression quantification and baySeq for differential expression testing.

cxb

[kstaplet]

Dessie-

I'm encountering the same problem you had, with my job getting dumped at the same "Processing Locus chr1:68024242-68025763" locus. Did you ever find a solution?

[GenoMax]

Dessie-

I'm encountering the same problem you had, with my job getting dumped at the same "Processing Locus chr1:68024242-68025763" locus. Did you ever find a solution?

Are you running this on a cluster or a local computer? How much RAM is available for this job?

[kstaplet]

I'm running on a cluster. I've used Cufflinks and Cuffdiff in the past and not had these problems.

I initially used Tophat to generate alignments using a reference transcriptome GTF, then used Cuffdiff using the same reference GTF. It took over a week to run, but finally completed successfully.

However, I didn't get many differentially expressed genes, so I reran Cuffdiff using a different reference transcriptome, hoping for more results. After taking almost a week "Calculating preliminary abundance estimates", it gave the error

Testing for differential expression and regulation in locus.
> Processing Locus chr1:68024242-68025763 [ ] 2%/opt/sge-6_2u4/default/spool/sunnode17/job_scripts/483892: line 11: 17962 Segmentation fault (core dumped)

Could it be because I tried using a different reference GTF for Cuffdiff and Tophat? Regardless, I'm still confused why Cuffdiff is taking so long. I'm also trying to identify novel transcripts with Cufflinks, and those are taking over a week per sample as well. I saw the suggestions for using a mask file (good resource here), so I'm going to try including that and see if that's my problem.

[GenoMax]

How big is the data file? A week per run seems too long (is that the actual time the job was running?) unless your data file is hundreds of GB in size.

[biocomputer]

Hi blancha,

Just an update on my efforts, I haven't solved the mysterious problem yet.

First, I did check with the administrator and I am getting 8 processors and I am not exceeding the memory limit.

Second, I did check on IGV several of my 12 samples, I looked at the bam files ( asd.bam files which contain all mapped reads from tophat accepted_hits.bam + properly tagged unmapped reads and is sorted and duplicate tagged). In the problematic chr1:68024242-68025896 locus I see no reads aligned whatsoever. The IGV has not listed a refseq gene in this locus and the nearest Erbb4 gene has many reads aligned to it. The UCSC genome browser has listed a predicted gene from Ensemble 75.

Third, I tried running cuffquant on one of my samples with Cufflinks 2.2.1 and it was running quicker than cuffdiff but stopped at the exactly same locus with the same segmentation fault, core dumped error message.

I have also tried re-running cuffdiff with the accepted_hits.bam output from tophat and again the same error came up.

To be honest I am really desperate now and I just wish I could get my differentially expressed genes.

Any ideas?

I'm having pretty much the exact same problem. I'm getting a segfualt from all the Tuxedo tools I've tried (cufflinks, cuffdiff, cuffquant), I'm not exceeding memory, they're are few or no mapped reads in the region it is segfaulting at. In cufflinks for example it segfaults during the section "Re-estimating abundances with bias and multi-read correction" while all the previous sections complete successfully.