How do I identify reads that overhang the end of a reference?

[Graham Etherington]

Hi,
I have 76nt single-end reads and I want to map them to a reference and identify the reads that overhang the end of the reference.
What would be the best mapper to do the mapping with?
If I use BWA as the mapper, I can find reads that have up to about 9nt overhangs (by comparing the reads' unclipped start site to the length of the reference the read maps to), but what I'm looking for is some way to find reads that overhang by about 40 nts, i.e. the first 37nts of the read match the end of the reference, but (obviously) the last 40nts don't.
Do any next-gen mappers allow such mapping?

[maubp]

Well the SAM/BAM file format itself doesn't allow this directly (the start/end co-ordinates must be within the length of the reference), so any reads which did map off the ends would probably have to be represented with a large insert or soft clip at the end.

I'm not sure of any specialized tool to do this. What I would consider is trying to map the first and last halves of your reads to the start/end of the reference. i.e. Some scripting to prepare the modified input files, then reuse a mapper like BWA, and interpret the results accordingly. Fiddly - so I hope someone else has done the hard work already!

[strob]

Dear all,

I want to re-open this discussion as I have a similar question: how can I detect and extract overhanging reads, and this at both ends of my reference?
I have overlapping MiSeq reads which I merged, resulting in reads with an average length of 300 - 350 bp. I'm only interested in what we call hybrid reads: reads partly mapping with the borders of my reference.
Any advice on a mapping tool? Even use Blast?
Thanks for the input.

[Brian Bushnell]

strob,

If you run BBMap with default parameters, it will soft-clip ONLY reads that map off the ends of reference scaffolds. So, you would just have to filter the sam file for reads containing the "S" symbol in their cigar strings.

You can adjust the fraction of the read that must be on the reference with the "minratio" flag (default 0.56). I suggest something like this:

minratio=0.2 maxindel=0

...which will look for alignments with as little as 20% of the read over the reference and not look for alignments containing indels.