Converting (existing) genomic alignments to transcriptome coordinates

[Rob Patro]

Hi all,

  This post is in (cross-)reference to a question I've posted over on Biostars.  I'm asking this question here because there doesn't seem to be a separate "developer" user-group.  Basically, as I understand it, the new "quantmode" allows STAR to output alignments in both genomic and transcriptomic coordinates, which is very useful.  In fact, the question I reference above is looking for a tool that's capable of projecting genomic alignments onto a set of transcriptomic coordinates (given, say, a .bam file with the genomic alignments and a gtf file with the transcript annotations).  My question is, how difficult might it be to leverage the code already in STAR that maps between genomic and transcriptomic coordinates to build a tool that does this?  Obviously, one could just use the existing switch when generating new alignments.  However, it would also be useful to have a tool to convert genomic alignments (perhaps only readily available only as bam files) into transcriptomic coordinates.  Given that my current search turned up nothing along these lines, I wonder if digging into STAR a little bit might turn up some code that is useful toward this end.

Thanks!
Rob

[Alexander Dobin]

Hi Rob,

such a conversion starting from BAM files is in my plans, however, it's not very straightforward, since I am using STAR internal representation of alignments to make the conversion - so I would have to convert SAM into this internal representation. Also, I am using some STAR specific data structures for output and parameters control, so it's not going to be easy to separate the conversion code. I might be able to work on it towards the end of October.

Cheers

Alex

[Rob Patro]

Hi Alex,

  Thanks for the quick reply.  I would have imagined that the code to perform this operation was fairly embedded in STAR.  However, it's one of the few (maybe the only) tool I know with an internal representation of alignments that allows outputs to both genomic and transcriptomic coordinates, so I figured it would be the best place to look.  I understand that this is probably a low-priority feature request, but if you're able to implement this, that would be fantastic.  I'll be sure to keep my eye on it!

Thanks,

Rob

[Daniel Gerlach]

Dear both,

Are there any updates in the mean time? I would like to use RSEM on BAM files retrieved from GDC which have been mapped to the genome and thus contain genomic coordinates. RSEM however, requires a BAM file with transcriptome mappings as an input.

Daniel

[Alexander Dobin]

Hi Daniel,

this option is still not available, unfortunately, with not ETA.

You would need to remap the reads to get transcriptome alignments.

Cheers

Alex