Mapping to transcripts rather than to the genome in RNA Seq

[jhbadger]

I inherited an RNA-Seq project with the goal of looking at differential expression across different environmental conditions and species in protists (using SOLiD, although probably the question is more general), that originally used transcripts as the mapping target rather than the genomes, largely for technical rather than scientific reasons. When I inherited it, I moved to mapping to the genomes, but the results don't correlate as much as one would think to the previous results (based on the same reads) and the experimental collaborator likes the transcript based results better as they behave closer to what he expects. So, one possibility is to ignore the genomes and go back to the transcripts for mapping.

What I am asking is if there is any literature dealing with the difference between transcriptome and genome mapping for RNA-seq and are there any scientific reasons to support transcriptome mapping even if the genome is available? I have tried to do a literature search myself, but I haven't seen anything that addresses the issue. Does anyone have any insight on this issue?

Thanks,
Jonathan

[EricHaugen]

A short read that maps unambiguously to one location in one genome may match multiple locations equally well in another genome. Depending on how your alignment program handles multiply-mapping reads, a read with the same conserved sequence could contribute to a transcript count in one genome but not in another.
Mapping just to the transcripts, you would have fewer problems with differential unique-mappability confounding the cross-species comparisons.

It looks like this study dealt with that issue, mapping to genomes separately but restricting the comparison to unique regions:

http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3296765/

[jhbadger]

Thanks! This is very helpful.
Jonathan