HTSeq 0.6.1 Maximum alignment buffer size exceeded error

[dvanic]

Hi! I'm usg HTSeq v 0.6.1 to count reads mapping to features (the htseq-count script). I'm getting a

18700000 SAM alignment record pairs processed.
Error occured when processing SAM input (record #40455805 in file /scratch/me/accepted_hits.bam):
  Maximum alignment buffer size exceeded while pairing SAM alignments.
  [Exception type: ValueError, raised in __init__.py:671]

My accepted_hits.bam file is generated by mapping with STAR and then coordinate-sorted and converted to BAM using samtools.

I am using the htseq command

htseq-count --format=bam --order=pos --mode=union --idattr=gene_id --stranded=reverse --type=exon} $WORKDIR/accepted_hits.bam gencode.v19.gtf

What is wrong and how do I fix it? (Apart from running htseq on the sam file)?

Thanks in advance,
Darya

[Simon Anders]

The "--order=pos" option is new and may need a bit of fine-tuning. Perhaps the buffer is a bit small.

For now, maybe sort the file be read name (with "samtools sort -n") and then use "--order=name".

[markf]

I can confirm this happening:

...
37400000 SAM alignment record pairs processed.
37500000 SAM alignment record pairs processed.
Error occured when processing SAM input (line 78096327):
  Maximum alignment buffer size exceeded while pairing SAM alignments.
  [Exception type: ValueError, raised in __init__.py:671

[colindaven]

Same here with
a) BAM ordered by position
b) STAR alignments

[Bukowski]

Have hit the same problem with STAR aligner, and positional ordering.

12200000 SAM alignment record pairs processed.
Error occured when processing SAM input (record #27408270 in file alignments/S6.bam):
Maximum alignment buffer size exceeded while pairing SAM alignments.
[Exception type: ValueError, raised in __init__.py:671]

[mehaffeymg]

I get the same "Maximum alignment buffer size exceeded" whether running my bam file as position or name sorted. It gets farther with the name sorted file than the position sorted file (output size of 1 Gb vs. 600+ Mb respectively) but still fails. My input bam is 11Gb of around 142 million reads.

[Simon Anders]

When you used the name-sorted file, have you remembered to omit the "-r pos"?

[mehaffeymg]

Working now

When you used the name-sorted file, have you remembered to omit the "-r pos"?

Excellent catch. I reran without that flag and it worked just fine.

Thank you!

[sidderb]

As Simon reads these I just wanted to flag up that I have run into the same issue with a position sorted alignment from STAR as well.

I was so excited to see this feature in 0.6 as the position to name based ordering is an annoyingly long step in my workflow! Fingers crossed for a fix soon...

[dakl]

I get the same with reads aligned with star, and position sorted using samtools.

tools/samtools view /home/daniel.klevebring/cust001/tests/fullRNA/L8760/rnaseq/L8760T/L8760T.star.50.stranded.PE.bam 
| htseq-count  --stranded=reverse --mode=intersection-nonempty  --order=pos - /mnt/hds/proj/cust001/autoseq_genome/genes/genes_fixed.gtf > /home/daniel.kleveb
ring/cust001/tests/fullRNA/L8760/rnaseq/L8760T/L8760T.star.50.stranded.PE.htseq-count.txt

...
18900000 SAM alignment record pairs processed.
Error occured when processing SAM input (line 40808946):
  Maximum alignment buffer size exceeded while pairing SAM alignments.
  [Exception type: ValueError, raised in __init__.py:671] 

$ htseq-count |grep version
Public License v3. Part of the 'HTSeq' framework, version 0.6.1p1.

I don't always get it, and it runs with larger files without a problem. What's the best fix for this? Is there one?

[catbus]

It appears that this problem is not addressed in 0.6.1p2 either.

It would be extremely nice if there were a way to use htseq-count on positionally-sorted data (which is the default Tophat output). Maybe there could be an option that htseq-count could use to just increase the buffer size? I have 48 GB of RAM on this machine, so it is not possible that htseq-count could actually consume it all on one file.

[golharam]

I get this problem as well using tophat2 aligned BAM files, position sorted. I will revert to name sorting.

[dc3000]

Is there a way to increase the buffer size?

I got this error with a pos-sorted.bam file (3.2GB) created with tophat2

[xxx@dev-intel14-phi 20131017mRNA_gfCC]$ htseq-count --format=bam --stranded=no --order=pos \
> ~/RTSF/hiseq/20131017mRNA_gfCC/ms12cc2/accepted_hits.bam \
> ~/RTSF/hiseq/20131017mRNA_gfCC/genes.gtf \
> > ~/RTSF/hiseq/20131017mRNA_gfCC/htseq/ms12cc2.txt
100000 GFF lines processed.
...
16100000 SAM alignment record pairs processed.
16200000 SAM alignment record pairs processed.
Error occured when processing SAM input (record #35593187 in file /mnt/home/xxx/RTSF/hiseq/20131017mRNA_gfCC/ms12cc2/accepted_hits.bam):
  Maximum alignment buffer size exceeded while pairing SAM alignments.
  [Exception type: ValueError, raised in __init__.py:671]
[xxx@dev-intel14-phi 20131017mRNA_gfCC]$ Write failed: Broken pipe

[catbus]

Two options:

You can either re-sort your files lexographically/alphabetically and run htseq-counts with the lexigraphic order (in other words, NOT --order=pos)

or

You can switch to the much faster and generally more user friendly (easier options, easier output file format) program "featureCounts". Part of the "subread" package. http://subread.sourceforge.net/

I would switch to subread. htseq-counts is incredibly slow and very poor at handling large files without crashing.


[QUOTE=dc3000;161213]Is there a way to increase the buffer size?

I got this error with a pos-sorted.bam file (3.2GB) created with tophat2

[CODE]
[xxx@dev-intel14-phi 20131017mRNA_gfCC]$ htseq-count --format=bam --stranded=no --order=pos \

[catbus]

P.S. subread only takes about 10 minutes to install. It doesn't have any complicated dependencies. Once again: http://subread.sourceforge.net/