Cufflinks and GTF files... something strange happens

[spacup]

Hi all !

Something strange happens when I use Cufflinks with 2 different GTF files...
FPKM are very differents...
An example of Cufflinks output for a gene with Ensembl GTF file :

ENSG00000188157 - - ENSG00000188157 AGRN - chr1:955502-991496 - - 3.40838 2.02671 4.79006 OK

The same gene with UCSC GTF file :

AGRN - - AGRN AGRN TSS19857 chr1:955502-991499 - - 1.8592 1.52117 2.19724 OK

Genes locations are almost same, but FPKM are very very different...
And this difference is not the same for all genes...
Is it normal?

Thanks!

[mattia]

I suppose it could be an annotation problem: as you can see in figure in attach, for the same gene you can have different exons length (choosing different annotation). So, different transcript length (i.e. different annotation) = different FPKM.

[spacup]

Oh I see, so even if the size of the gene is similar, the number of exons being different, results are different...
So.... which one is better to make analysis ? Generally which one is used ???

It seems to me that using GTF files is thus too much hazardous and that Cufflinks is not the best tool for DE analysis.....

[mattia]

In my opinion the problem is not "how cufflinks (cuffdiff for DE) works": I'm using Cufflinks and it works very good.
I suggest to use always (either using Cufflinks and HTS-count, edgeR,....) just one annotation (i.e. GTF and also reference.fasta) : from alignment to DE. First step of a good analysis is to FIX an annotation and to keep it until the rest of analysis.

[Helba]

I suppose it could be an annotation problem: as you can see in figure in attach, for the same gene you can have different exons length (choosing different annotation). So, different transcript length (i.e. different annotation) = different FPKM.

Hello!
Need to perform annotation of new genes based on RnaSeq in a genome already annotated. For this precise masking regions that are already annotated. How to do this??
"cufflinks-M transcripts.gtf Controle.sam"
This command missing something?

[spacup]

Ok, thanks a lot for your answers.

[sdriscoll]

I think in your screenshot there the gene tracks are condensed. One possible difference is also because Ensemble provides at least one splice variant of this gene while the UCSC annotation shows it as a single isoform gene. That most certainly will have an impact on how cufflinks assigns expression even at the locus level.