Run cufflinks with or without annotation?

[reut]

Hi all
From what I see in this forum, the common pipeline for Cufflinks is to run it without a reference annotation and supply the annotation only to cuffcompare.

If I run cufflinks with annotation, it will not build novel transcripts, but are there any additional considerations?
i.e. If I don't care about novel transcripts, is it better/recommended for me to run cufflinks with annotation?

thanks in advance,
Reut

[DZhang]

Dear Reut,

The latest version of Cufflinks allow you to supply an annotation file and with two options: consider the reads that only compatible with the annotation, or building novel transcripts in addition to the annotation.

Check the online manual for details.

Douglas
www.contigexpress.com

[kopi-o]

I would think it should be better to run it with an annotation if you are not interested in novel transcripts.

[pbluescript]

I run it with a reference. In fact, unless you have very good coverage, I think running it with a reference is a better way to go. Unless you have split or paired end reads linking every part of a gene together, running Cufflinks without a reference annotation could cause it to split a single gene into multiple parts. Avoiding an artifact like that is more important to me than using Cufflinks to identify novel transcripts. There are other tools to use for that.

[reut]

Thank you everyone and thank you pbluescript, what you are describing is exactly what I see - when running Cufflinks without reference I see a single gene split into multiple parts.
So I will run Cufflinks only with reference.

[Hunny]

Originally Posted by reut

Thank you everyone and thank you pbluescript, what you are describing is exactly what I see - when running Cufflinks without reference I see a single gene split into multiple parts.
So I will run Cufflinks only with reference.

I have a question here.

As mentioned in the mannul of cufflinks, "Output will include all reference transcripts as well as any novel genes and isoforms that are assembled." is for the "-g/--GTF-guide" option.

My question is if the predicted transcripts don't include the
reference transcripts, how could I differentiate the predicted transcripts from the output of cufflinks using the -g option? It seems I could only see the novel ones, compared to the reference.

And how could you prove the "multiple parts" are artifacts and remove them?

Cheers,

[polyatail]

My question is if the predicted transcripts don't include the reference transcripts, how could I differentiate the predicted transcripts from the output of cufflinks using the -g option?

From the manual, it sounds like it will output all provided reference transcripts in addition to any novel isoforms. If I'm understanding your question correctly, it sounds like you'd be able to identify novel isoforms by subtracting the reference GTF file you provide from the Cufflinks output GTF.

And how could you prove the "multiple parts" are artifacts and remove them?

If you load the Cufflinks output GTF file into say, the UCSC genome browser, and look at your favorite gene, you'll see that a multi-exonic transcript appears as several, smaller "fragment" transcripts--some single exon, some just a few exons--but all incomplete as compared to the reference. In theory, you could strip all of these by comparison to the reference set of transcripts, but then you'd eliminate the novel isoform discovery function of Cufflinks, and might as well run with that function disabled. The conclusion reut and pbluescript reached is that, when run with the --GTF-guide option, most of these spurious transcripts will be eliminated while still allowing novel isoform discovery.

[polyatail]

Has anyone found any differences in Cufflinks output when providing a reference via -G at the TopHat step? With or without also providing a reference to Cufflinks?