StringTie discovers less novel isoforms when reference annotation is provided

Question

I am analyzing PacBio IsoSeq data and I am using StringTie to assemble transcripts. I noticed the difference in the output files when I add a reference genome annotation (-G option in StringTie). In general I have less transcripts assembled when I give as imput reference genome annotation and I was wondering why. I know that the annotation of the genome is not the best (Borelian orangutang) and I would rather trust the data rather than the annotation provided.

I am attaching an example for one of the genes (IGV): the first track is reference annotation; the second is StringTie output when no reference annotation is provided; the third is StringTie output when reference annotation is provided (in the third case two novel assembled isoforms are missing)

Answer 1

Not very familiar with what StringTie is doing, but in general reference annotations are helpful in providing a prior for what the tool expects.

I would suggest looking at the manual and using some of the provided options and outputs to probe the coverage of the various transcript species (I believe these are in the "attributes" section of the GTF output).

If the transcripts that are missing when you use -G have a coverage of just a couple of reads, they are likely to be less interesting. On the other hand, if they have relatively high coverage, then you have more reason to be skeptical of the existing annotation.

My guess is that the tool is more heavily weighting existing transcripts based on the annotation, and therefore excluding the extra transcripts as being below some threshold.

I'd recommend consulting the "recommended workflow" in the manual that supposedly allows for novel isoforms. Ensure that you are using this workflow and not the more streamlined workflow, which does not allow for novel isoforms.

See also the comment suggesting comparing to other IsoSeq-oriented tools.