Tuxedo pipeline: cufflinks with -G option vs cuffdiff with reference

[asier_gonzalez]

Dear all,

I have been analysing RNAseq experiments using the default tuxedo pipeline for about a year, i.e., TopHat2 + cufflinks + cuffmerge + cuffdiff2 and one thing I have never understood is the value that running cufflinks with the -G/--GTF option adds to the analysis.

Considering that cufflinks acts as a transcript assembler, with the -G option it does not search for novel transcripts, which means that the output GTF contains the same transcripts as the original reference. Therefore, cuffmerge will merge as many reference annotation files as available samples which will produce, again, the reference annotation. The only difference is that if cuffmerge is run including the fasta reference, it will add some attributes that allow cuffdiff to test for differential expression at the TSS and CDS level.

Considering all this, my question is whether there is any big difference between running the whole pipeline without searching for novel isoforms (cufflinks -G) and running directly cuffdiff with the reference annotation directly. I must say that I have tried this in the past and that the gene level results where slightly different, but most importantly, the second option is much faster than the whole pipeline.

Any input will be highly appreciated.

Thanks,

Asier

[GenoMax]

Just like you discovered some folks are not interested in finding new transcripts/splice isoforms and are happy to use the "known transcriptome" for the expression analysis. So doing a special one-time TopHat run with the -G option to create a transcriptome specific index subset of genome speeds the analysis process up.

[asier_gonzalez]

Thank you GenoMax,

So would you say there are any differences between running only TopHat + CuffDiff and TopHat + Cufflinks (-G) + Cuffmerge + CuffDiff if you only want to stick to the annotated transcriptome?

I am asking because I work with wheat, which has a 16 Gb genome (12 Gb latest reference) and skipping cufflinks and cuffmerge means that you can save up to one week depending on the experiment.

Asier

[Brian Bushnell]

If you are not interested in novel isoforms, but you are interested in saving time, it's much faster to skip TopHat and map to the transcriptome only, using a different aligner. Or even better, an alignment-free tool such as Seal or Sailfish.

[asier_gonzalez]

Thanks for your input Brian,

I accept that using the transcriptome as the reference is much quicker and the tools like Sailfish and Kallisto are amazing in that sense. However, mapping to the transcritome means that if a gene is not annotated properly, say it is a partial gene, a read could map beyond its boundaries in the genome but would not map in the transcriptome.

Having saying this, I have used the whole Tuxedo pipeline searching for new isoforms, but cufflinks is not flawless. One of the problems it has is that it may merge genes thar are close.

Considering all these, I still try to understand if running cufflinks with the -G option has any advantages over skipping it.

Thanks again,

Asier