cufflinks / cuffcompare does not produce p_id

[dariober]

Hello,

My first post here...!

I'm trying to use cuffcompare to produce a combined GTF file (stdout.combined.gtf) to feed to cuffdiff for further analysis. Cuffcompare should produce such GTF with added attributes tss_id and p_id. However, while my stdout.combined.gtf does have tss_ids it has no p_ids (even if the input file has CDS attributes). How is it?
I guess I can put the p_ids myself, but I'd like to understand what I'm doing wrong.

As a test, I ran cuffcompare using the pig GTF from ensembl against itself (following a trick from Cole Trapnell I found in another post) and indeed I couldn't 'see' any p_id in the output. See script below to reproduce this test.
Why I don't get p_ids?! Am I missing something very obvious?

I'm using version cufflinks-0.8.2.Linux_x86_64

Thanks in advance for any help!
Dario

Here's the little code:
# -----------------[ Cut here ]---------------------

## Download and unzip pig GTF
wget ftp://ftp.ensembl.org/pub/current_gt...ofa9.57.gtf.gz
gunzip Sus_scrofa.Sscrofa9.57.gtf.gz

## Run cuffcompare
/exports/work/vet_roslin_nextgen/dario/cufflink/current/cuffcompare -r Sus_scrofa.Sscrofa9.57.gtf Sus_scrofa.Sscrofa9.57.gtf Sus_scrofa.Sscrofa9.57.gtf

## No p_ids!!
grep p_id stdout.combined.gtf
==> no hits

# ----------------------[ Cut here ]-----------------

[Cole Trapnell]

The trick I posted apparently doesn't always work - we know about this bug and will address it in an upcoming release

[dariober]

The trick I posted apparently doesn't always work - we know about this bug and will address it in an upcoming release

Hi Cole et al.,

Thanks for taking time to reply.
Is the bug due to some anomalies in the GTF file? If so, would it be possible to post a few lines from a GTF that works? A final question if you can answer... even if this bug occurs, are the other results from cuffdiff reliable? (I'm not interested in running the reference GTF against itself, that was just for testing purposes).

Dario

[Cole Trapnell]

We think they're reliable

The bug I mentioned is just in cuffcompare's ability to attach p_id attributes automatically. If the p_id attributes are missing, Cuffdiff will simply not perform a CDS-level differential analysis. It will look for changes in splicing and TSS preference (and of course gene and isoform-level expression), but it won't look for changes in coding output.

[rocksd]

We think they're reliable

The bug I mentioned is just in cuffcompare's ability to attach p_id attributes automatically. If the p_id attributes are missing, Cuffdiff will simply not perform a CDS-level differential analysis. It will look for changes in splicing and TSS preference (and of course gene and isoform-level expression), but it won't look for changes in coding output.

Cole,

Did you fix this bug?
I am running a Cuffdiff on two pig samples and ended up with Segment error. I then checked the combined_transcript.gtf file generated in Cuffcompare but did not find the p_id. By the way, I am using the cufflinks .8.4 version released 16/7/2010

Thanks a million!

jhuang