cufflinks: parameter -g

[mrfox]

hi all,
I am using the cufflinks->cuffmerge->cuffdiff pipeline to assemble transcripts (known and novel) and perform differential analysis. I am not sure if I should use the parameter -g to "guide" the assembly using an annotation.

I am not sure how much difference using or not this parameter if I want to identify novel transcripts. Maybe this parameter should not be used in thecufflinks step but in the cuffmerge step instead.

Any advice will be appreciated.

[arvid]

It depends on how mature the annotation is, and how close you can expect your samples to be. For me, I usually see much better results, regarding finding new transcripts (and avoiding false positives), when using reference-guided mapping (with GSNAP and TopHat) and assembly (with Cufflinks), but I'm working on highly homozygous plant samples, so your mileage may vary.
To me, however, the alignment step seems to be more sensitive to this, however - there I seem to avoid most weird alignments when using the known transcriptome (with its splice sites) as a guide... The assembly step, to me, is less sensitive.