bs-seq in colorspace

[brentp]

there is some discussion of this in another thread.

i'm considering doing the following for bs-treated reads in colorspace:

+ convert C to T in reference (and for reverse-complement of reference)
+ for reads:

- convert to nucleotide space
- convert C to T
- convert back to colorspace

+ use aligner (bowtie/bfast) to do mappings
+ ... do usual stuff to calculate per-base methylation


what's wrong with this approach? or how could it be improved for colorspace? is this valid with a subset of reads, so that i could align some this way and then use, for example SOCS (which is very computationally intensive) for unmapped reads?

thanks for any ideas.

[mrawlins]

Converting to nucleotide space is generally a bad idea. Converting back will sometimes alleviate this problem, but I don't think that's true in this case.

A single mismatch in colorspace will change the identifications for all subsequent positions. That will mean:
1) some bases that were C were not called correctly, so not changed to T. The colorspace representation doesn't change even though it should.
2) some bases that weren't C were incorrectly called as C, so they were changed to T. The colorspace representation changes even though it shouldn't.

It might be possible to find color transitions that imply a base is C and change those to the corresponding transition for base T, but I think that would miss a few cases, and hence a lot of reads.

I could be wrong on my assessment here, but it seems like it would produce quite a few false positives and false negatives. I don't have any better ideas, though.

It seems that the new ECC stuff from ABI would alleviate a lot of this issue, as it allows trivial conversion to base-space. That assumes, though, that you haven't run the samples yet and have available either the equipment or a collaborator/service that has the latest equipment, as it doesn't work on SOLiD 4 or older.