BWA 0.6.1 fail to insert infer isize: weird pairing + Segmentation Fault

[attilav]

Hi all!

Using the command: bwa sampe refseq.fasta seq_1.aln.sai seq_2.aln.sai seq_1.fastq.gz seq_2.fastq.gz > output.sam , sometimes I get the error that you can see in the title. It looks like this:

[bwa_sai2sam_pe_core] 5505024 sequences have been processed.
[bwa_sai2sam_pe_core] convert to sequence coordinate...
[infer_isize] (25, 50, 75) percentile: (410, 437, 469)
[infer_isize] fail to infer insert size: weird pairing
[bwa_sai2sam_pe_core] time elapses: 0.82 sec
[bwa_sai2sam_pe_core] changing coordinates of 643 alignments.
[bwa_sai2sam_pe_core] align unmapped mate...
Segmentation fault

I did several very similar alignment tasks with BWA, most of them goes through fine, but with a few fastq files this happens. My data is from the 1000genomes, so that shouldn't be the problem I guess.
Please let me know, if you guys have an idea what's causing this!

Yours,
Attila Vikor

[swbarnes2]

The simplest thing is to redo the alignment. If you mistyped a file name somewhere, you might get that result.

Also, try making the single end .sams separately, and spot check a few pairs of reads, see if thei rcoordiantes and relative orientation make sense. Weird pairing might mean that the reads don't run into each other like they should.

[Richard Finney]

Is your genome index made with bwa 0.6.1?
How much memory do you have ?
Is the error consistent (reproduceable)?

[attilav]

I made the genome index with bwa 0.6.1.
I have 2GB RAM on this computer, which is not optimal, but the segfaults only happen, if there is a weird pairing before.
Its consistent.

I will align the files separately, and see if something comes up.

[Richard Finney]

I think you're going to need more that 2GB.
Try one with 8GB on the problem samples.

[attilav]

Apparently the problem was some of the files getting corrupted during downloading. Sorry to have wasted your time with this... Guess I have learned a lesson to always check the MD5 sums when downloading huge files from the internet.