cuffdiff: making sense of gene_id, p_id, transcript_id and tss_id

[Aurelien Mazurie]

Greetings to Cufflinks users,
I am trying to interpret outputs from cuffcompare and cuffdiff to answer the following questions: is there some alternative splicing going on in an experiment (3 time points), and if yes, what is the absolute (or relative) abundance of the alternative transcripts and of the genes themselves (by summing up the abundance of their isoforms, or by considering only the major transcript).

Despite my reading of the Trapnell et al. 2010 paper about Cufflinks I am still unsure how to read the files cuffcompare and cuffdiff are generating.

For example, I understand the various identifiers (gene_id, p_id, transcript_id, tss_id) used by these tools track the objects I am interested in; however, I don't understand how they relate to each other. The long threads on SEQanswers about identifier missing or appearing when one forgot options (such as -s for cuffcompare) only add to the confusion.

Hence, can somebody come up with a simple explanation about how these identifiers relate to each other? I was hoping to find some figure in the Cufflinks documentation, but couldn't.

Out of curiosity I wrote a script to evaluate the cardinality of the mapping between each of these identifiers. E.g., if two objects in the combined.gtf file produced by cuffdiff have the same gene_id but two p_id then the mapping from gene_id to p_id is 1 to 2. Here is the result I obtain with my samples (all.combined.gtf file produced by CuffDiff):

gene_id -> p_id
1: 10019 2: 3
gene_id -> transcript_id
1: 10119 2: 5
gene_id -> tss_id
1: 4547 2: 2
p_id -> gene_id
1: 10025
p_id -> transcript_id
1: 10025
p_id -> tss_id
1: 4551
transcript_id -> gene_id
1: 10129
transcript_id -> p_id
1: 10025
transcript_id -> tss_id
1: 4551
tss_id -> gene_id
1: 4551
tss_id -> p_id
1: 4551
tss_id -> transcript_id
1: 4551

The number before the ':' is the cardinality of the mapping and the number after it is how many of these do we find in the file. Hence, for example, there are 3 cases where one gene_id is linked to two p_id.

As you can see, it seems that most identifiers map to each other with a 1:1 mapping, which does not help me.

EDIT

I performed the same kind of mapping analysis on various identifiers of the isoforms.fpkm_tracking file produced by cuffdiff, which appear to be the one file I am interested in. Here is the result:

gene_short_name -> nearest_ref_id
1: 9884 2: 47 3: 1 4: 1
gene_short_name -> tss_id
1: 4491 2: 14 4: 1
nearest_ref_id -> gene_short_name
1: 9985
nearest_ref_id -> tss_id
1: 4551
tss_id -> gene_short_name
1: 4523
tss_id -> nearest_ref_id
1: 4551

Is it right to interpret that I indeed have 14 genes with two isoforms (i.e., splicing variant), and one gene with four isoforms? tss standing for transcription starting sites, I would think that some isoforms have the same tss_id but different exons. How are these second type of isoforms reported?

Best,
Aurelien

[poisson200]

Dear Aurelien,
Your post is very interesting and I would, too, like to understand fully the output of cufflinks/compare/diff. The one thing that caught my eye is that you have got p_id's in your output. Is that right?

I have not managed so far to do this with hg19/Refseq_gtf data. It does not help your question but it would be great to know the species and the annotation files you used to get your data.

Thank you for any advice,

Kind regards,

John.

[adrian]

Hi Aurelien,

In the edit you mentioned that isoforms.fpkm_tracking file is the one you are looking for. how does this one file answer your question:

"is there some alternative splicing going on in an experiment (3 time points), and if yes, what is the absolute (or relative) abundance of the alternative transcripts and of the genes themselves (by summing up the abundance of their isoforms, or by considering only the major transcript)."


I read and re-read Trapnell et al. paper and I am sifting through each word of cufflinks manual, I am still lost.

I successfully ran cufflinks, cuffcompare and cuffdiff. I just dont know how to interpret the tables.

the questions I am interested in (copying your word):

is there some alternative splicing going on in an experiment between disease and normal tissue given 3 regions of brain.

frontal lobe ( disease vs normal)

temporal lobe (disease vs normal)

whole brain (disease vs normal)

If yes, what is the absolute (or relative) abundance of the alternative transcripts and of these genes themselves (by summing up the abundance of their isoforms, or by considering only the major transcript).


Thanks