What determines which tests are performed in Cuffdiff? Only two types of tests were performed in our Cuffdiff run

Question

[tunl]

Our recent Cuffdiff runs only performed “isoform-level transcription difference tests” and “gene-level transcription difference tests” for some reason (other tests were not performed), as follows:

Performed 84687 isoform-level transcription difference tests

Performed 0 tss-level transcription difference tests

Performed 22374 gene-level transcription difference tests

Performed 0 CDS-level transcription difference tests

Performed 0 splicing tests

Performed 0 promoter preference tests

Performing 0 relative CDS output tests

Previously when we ran a different set of RNA-seq data on Cuffdiff, all types of tests on this list were performed.

So I am wondering what determines which tests on this list are performed in Cuffdiff?

We do not specify this in running Cuffdiff though.

Any ideas and advice would be greatly appreciated.

Thank you very much in advance!

[microfuge]

I think that should depend on the gff annotation file supplied. If it has no cds entries no cds test is performed and likewise.

[tunl]

Thanks so much!

We use GENCODE gtf this time (last time we used UCSC gtf). So GENCODE gtf seems to have less entries than UCSC gtf (I wonder why this is the case).

How important are tss-level, cds-level, splicing, and promoter preference tests?

Thanks a lot!

[microfuge]

Am Glad. I did not find much use for tss or splicing unless the coverage is ultra high and the biological question warrants that kind of analysis.

[tunl]

Thank you very much!

[Satyajeet Khare]

I'm not sure about CDS, but Splicing and Promoter tests are performed only if you have 3 replicates (using default settings). You can perform Promoter and Splicing test with less number of replicates (default 3 can be lowered down) but it produces less number of events.

[tunl]

Thanks a lot!

We actually have 3 replicates for each condition, but Splicing and Promoter tests were not performed for some reason.

We didn’t use “-b” option, since the bias correction algorithm makes the “Testing for differential gene expression” step extremely slow (2-3% per day). In the gene_exp.diff, we only got ~50 differentially expressed genes, which is a small number. I’m not sure if this is related to that other tests were not performed?

Also, may "no -b option" be possibly a cause of that the number of differentially expressed genes is small?

Thank you very much!

[Satyajeet Khare]

This suggestion is not exactly "scientific" but try running the command again :) I have a feeling that it has happened to me in past that sometimes the cuffdiff failed to complete the analysis.

Also, can you please provide your command lines for the entire analysis?