Paired read names / SAM qname format

[misko]

Hi everyone,
I am trying to figure out what possible paired read name formats are in use and good to support for open source alignment programs. The most common paired read name format i encounter is something like the following,
>READ_A/1
AAAAACCCCCCC
>READ_A/2
TTTTTGGGGGAA
Is this the only format used by different sequencing platforms? Are there other formats such as,
>READ_B:1
AAAAACCCCCCC
>READ_B:2
TTTTTGGGGGAA
Also with respect to SAM format, does anyone know what the proper query name for these reads should be?
Most aligners that I have tried seem to agree that for paired reads that end in "/1" or "/2", these last two characters are truncated to get the SAM query name.
For example the first read pair would be reported as,
READ_A
But the second read pair would have two query names,
READ_B:1 and READ_B:2 in the SAM file
I was thinking that maybe the safest approach would be to find the largest common prefix and truncate there to get the pair read name. But I am not sure if this will always work, for example, if the read names are,
>READ_C_Z12411/1
AAAAACCCCCCC
>READ_C_Z12516/2
TTTTTGGGGGAA
The pair read names based on largest common prefix would be,
READ_C_Z12
Where as most aligners would report two separate names as,
READ_C_Z12411 and READ_Z12516
Thanks for the help!

Misko

[jkbonfield]

As you've observed, Sam QNAME isn't what traditional assemblers and viewers refer to as a read name. Instead it's really the template name - in your example qname is READ_A or READ_B. I think the automatic detection of /1 and /2 to trim down a read-name to a qname is there simply because /1 and /2 was the MAQ convention. Ideally no one should rely on this convention now.

A common prefix is prone to error too. When dealing with fastq there's a huge variety of options:

- 1 fastq line per template, with an implicit assumption we chop in half to get both reads (the original illumina format IIRC)

- 1 fastq file with fwd and rev reads alternating, so they always come in pairs.

- 2 fastq files with each end stored in its own file.

- 1 fastq file for single-ended data. This obviously causes confusion with the first two conventions.

Every aligner seems to want something marginally different, and fastq frankly is hopeless as a format for embedding such meta data so it's tricky for tools to work out which read layout you use without either explicit command line options or trying to "be clever". In my experience that nearly always ends up shooting you in the foot sooner or later.

As for SAM. I'd like to see the true read suffix (or entire name if inappropriate) having a standard auxillary key:value tag to go from template names back to the original read names. Adding that and the ability to have more than 2 reads per qname and we gain the ability to use SAM for mixed assemblies with capillary finishing reads in it.

James

[misko]

Thanks for the answer, this helps clarify