Circular reference mapping

[arvid]

Since there seems to be no read aligner handling circular references (at least which I could find), what are typical approaches you would take? I'm working with RNA-Seq data from chloroplasts and mitochondria.

I can think of simple ways, like concatenating the end parts (including pieces longer than the largest expected fragments), however I want to know what typical problems you face when taking such a route. I can think of a number or theoretical problems, but don't know how many of these actually turn out to cause headaches in the downstream analysis.

Or is someone out there working on circular reference addition to any of the popular aligners?

Cheers,
Samuel

[alexdobin]

I think concatenating the end parts is the way to go. I do not see any obvious way to do it more elegantly. Such a problem calls for something like periodic boundary conditions, but I cannot see how it could work with sequence search algorithms.

In terms of downstream analysis here are a couple of thoughts.
You would have to filter out alignments that map entirely into the concatenated parts. This filtering has to be done before you decide which alignments are multi-mappers, otherwise you may loose unique mappers that could, for example, map to the very beginning of the genome and also to the concatenated part at the end. To reduce this effect you may want to concatenate exactly (mateLength-1) bases.