Cufflinks -g option yields many 0 FPKM transcripts and annotates them interestingly

[pettervikman]

Hi

I've just started using the -g option in the newest version of cufflinks. I've found that it then creates a whole bunch of transcripts that are specified as 0 in the FPKM, as well as in high low conf. FPKM columns. It's easy to filter this after but I find that it's interesting that they are created initially.

Second I've found that when looking in the GENES file the transcripts are tagged as OK or LOWCOVERAGE. Here transcripts with a low-conf FPKM of 0 and any of the other FPKM fields above 0 are flagged as lowcoverage while transcripts with FPKM of 0 and all fields > 0 are OK. Once again it's easy to change the OK to not expressed or similar but it would be nice to not having to do this.

So my questions are 1, has anybody else seen this and 2, have anyone information regarding how "true" the prescence of transcripts are?

Sincerely

/Petter

[jhb1980]

Hi Petter,

I'm still pretty new to this all, so if anyone spots a wrong statement below, please feel free to point it out.

As for question 1, I think this is an inherent property of the -g (RABT) option, which tiles reference transcripts with faux reads wether the transcript is expressed in your sample or not (see http://cufflinks.cbcb.umd.edu/howitworks#hrga ). If you take your output through Cuffcompare, the "-R" option should remove any reference *.gtf transcript not expressed.

As for question 2, I don't think this is a trivial one and pretty much an own topic of research. The best ways (from a biological point of view) in my opinion would be to see if a) the transcript expression / (structure) is reproducible between replicates; and b) if the transcript is still detected when using higher mapping stringency (mismatch allowance, multimapper limit), i.e. "playing around" with the input parameters of the Mapper and Assembler. The first point of course raises your invoice considerable .

[pettervikman]

Hi

Thanks for the reply. I've realised that the output annotation is a bit unprecise, or I don't understand the labelling. Transcripts with 0 coverage and 0 in all positions except the highFPKM are labelled OK for example.

Anyhow, I have 48 samples and I've decided to filterout anything that is not present in at least two samples. After that I'll see what I have left.

/Petter