BWA mem paired end read name errors

Question

[dp]

When using the latest version (0.7.17) of bwa mem it seems like paired-end reads (split in 2 files) must have names like read1/1 and read1/2 (using the "/" character), while an older version I used (0.7.5) also allows names like read1_1 and read1_2 or read1.1 and read1.2 .

Why is this the case in the latest version? What is the most up to date version that supports these different naming conventions? Is there a way to prevent the most recent bwa mem from erroring out on reads that use these naming conventions?

[Pierre Lindenbaum]

as far as I can see, the code trimming the fastq is still here: https://github.com/lh3/bwa/blob/master/bwaseqio.c#L209

    { // trim /[12]$
        int t = strlen(p->name);
        if (t > 2 && p->name[t-2] == '/' && (p->name[t-1] == '1' || p->name[t-1] == '2')) p->name[t-2] = '\0';
    }

[dp]

Thanks for the quick response. I'm not sure I completely understood this code, but it looks like it specifically is trimming on the "/" character, so this won't happen on a "." or "_" character.

It seems that some read files use the "." or "_" instead of "/", and I am able to run bwa mem v0.7.5 on them. But in v0.7.17 I get the error that "paired reads have different names" since they are not properly trimmed.

I'm not sure why all the options can't be supported in the latest release.

[Pierre Lindenbaum]

Thanks for the quick response. I'm not sure I completely understood this code, but it looks like it specifically is trimming on the "/" character, so this won't happen on a "." or "_" character.

exactly

[Pierre Lindenbaum]

The code was the same in v0.7.5:

https://github.com/lh3/bwa/blob/ef18cb91cb2211250898d2ec94203b13328e4a6a/bwaseqio.c#L209

[dp]

Maybe there is somewhere else in the old version that handles the other formats?

[dp]

Or rather - maybe the old version doesn't have a check for whether the last part of the name is the same?

[John Marshall]

The real question is what program or instrument has produced these reads with unusual names like read1_1 or read1.1.

[dp]

It's Illumina HiSeq 2000