Segmentation fault

[Fujun Qin]

Hi, Alex,

I generated a genome(Actually it is a fusion gene)  successfully.  Then I ran mapping,  but I saw bellows:

__________________

-bash-4.1$ /net/midtier19/vol80/hlilab/software/STAR_2.3.0e.Linux_x86_64_static/STAR --runMode genomeGenerate --genomeDir /net/midtier19/vol80/hlilab/SLC-ELK-project/download/SRA/STAR/1024/ --genomeFastaFiles 16cis-SAGE.fasta --runThreadN 2Jan 14 13:44:47 ..... Started STAR runJan 14 13:44:47 ... Starting to generate Genome files

Jan 14 13:44:47 ... starting to sort  Suffix Array. This may take a long time...

Jan 14 13:44:47 ... sorting Suffix Array chunks and saving them to disk...

Jan 14 13:44:47 ... loading chunks from disk, packing SA...

Jan 14 13:44:47 ... writing Suffix Array to disk ...

Jan 14 13:44:47 ... Finished generating suffix array

Jan 14 13:44:47 ... starting to generate Suffix Array index...

Jan 14 13:44:55 ... writing SAindex to disk

Jan 14 13:45:12 ..... Finished successfully

-bash-4.1$ /net/midtier19/vol80/hlilab/software/STAR_2.3.0e.Linux_x86_64_static/STAR --genomeDir /net/midtier19/vol80/hlilab/SLC-ELK-project/download/SRA/STAR/1024 --readFilesIn /net/midtier19/vol80/hlilab/SLC-ELK-project/download/SRA/STAR/1024/1024/ERR031024_1.fastq /net/midtier19/vol80/hlilab/SLC-ELK-project/download/SRA/STAR/1024/1024/ERR031024_2.fastq --runThreadN 2

Jan 14 13:45:35 ..... Started STAR run

Jan 14 13:45:40 ..... Started mapping

Segmentation fault

_______________________________

if I mapped another genome, it worked well.

do you know what 's wrong with it ?

Thanks,

Fujun

[Alexander Dobin]

Hi Fujun,

I think this is caused by the reference being too small. What is the size of your reference? 

At the genome generation step, please try to reduce the value of --genomeSAindexNbases, for a 10kb reference I would recommend 6, and then re-run the mapping step.

Generally, --genomeSAindexNbases  needs to be scaled with the genome length, as ~min(14,log2(ReferenceLength)/2 - 1)

Cheers

Alex

[Fujun Qin]

Thanks, Alex.

But I am new user. I wonder how to change the genomeSAindexNbases.  

Thanks,

Fujun

[Fujun Qin]

the size is around 3kb.

On Thursday, January 16, 2014 6:02:36 PM UTC-5, Alexander Dobin wrote:

[Alexander Dobin]

Hi Fujun,

simply use  "--genomeSAindexNbases 5" in addition to your other STAR parameters.

Also, consider using the latest patch ftp://ftp2.cshl.edu/gingeraslab/tracks/STARrelease/Alpha/STAR_2.3.1x.tgz (the static version STARstatic is inside the source directory).

Cheers

Alex

[Fujun Qin]

Hi,Alex, 

I used the new version STAR, I also adjust the  "--genomeSAindexNbases " based on the  ~min(14,log2(ReferenceLength)/2 - 1).  But this time the it was "killed ".  I don't know what is the problem.

Thanks,

Fujun

[Yohann Nédélec]

Hi,

I also have a segmentation fault with STAR_2.3.0e when aligning reads to the human genome.
I both tested --genomeSAindexNbases at 14 (default) and 2 when generating the genome.

The two tests returned a segmentation fault after creating a SAM file of 1.8T.

They both crashed respectively at the 6083677884 and 6080658025 read, maybe that little difference is due to the --genomeSAindexNbases parameter.

We like to create our own junction file, so we cat reads from 180 samples for a total of ~6e9 reads.

It seems that the mapping is crashing at the very end.

Do you have any idea ? Is there something I can provide to help ?

Thanks,

Yohann

[Alexander Dobin]

Hi Fujun,

What was the actual value you used for --genomeSAindexNbases? You can try to go down even a bit more.

If a code gets killed like that most, likely it's because you did not have enough RAM to run this process.

Please try to monitor memory usage with top when you start mapping.

This is a bit strange since your genome appears to be very small, it took only a few seconds to generate it.

If nothing works, please send me the Log.out file from the killed run.

Cheers

Alex

[Alexander Dobin]

Hi Yohann,

this is likely a different problem from the above.

Please try the following:

Use the latest STAR patch:

ftp://ftp2.cshl.edu/gingeraslab/tracks/STARrelease/Alpha/STAR_2.3.1y.tgz

Increase --limitOutSJcollapsed to 2000000 or 5000000 or 10000000.

I suspect the following problem. By default STAR allocates space for 1,000,000 collapsed junctions, and if you have billions of reads you may have >1M junctions detected, which will overflow that space. The new patch should deal with this problem nicely instead of seg-fault.

Cheers

Alex

[Yohann Nédélec]

Hey Alex,

It worked, thanks !

Cheers,

Y.