Clarification on output-file contents: discovered_transcript_grouped_tpm.tsv VS transcript_grouped_tpm.tsv

[andysaiz]

Hi,

I ran IsoQuant 3.7.0 on bulk MAS-iso-seq Pacbio data but I have am a bit confused on a few aspects of the output files. I understand that IsoQuant runs in two stages, one that finds transcripts that match the reference, and the second that finds novel transcripts. However, several transcript IDs overlap between reference and discovered transcript outputs. I have these two files (among others) that I would like to use:

• OUT.transcript_grouped_tpm.tsv (REFERENCE)
• OUT.discovered_transcript_grouped_tpm.tsv (DISCOVERED)

In the DISCOVERED file, there are several ENST transcript IDs that are are also in the REFERENCE file. The TPM values are usually very similar but not exactly the same. Thus, should isoform IDs that overlap between the two be treated as two completely different isoforms even though they share the same transcript ID? Do you typically concatenate the outputs from the reference stage with the discovery stage for one complete set comprising novel + reference?

Say the DISCOVERED file contains 64,000 transcripts. Would it be incorrect to report that as 64,000 novel transcripts? I guess the fact that they are in the DISCOVERED file means that they must differ somehow from the reference, so they are truly novel, no?

[andrewprzh]

Dear @andysaiz

Thanks for the question - let me try to clarify this aspect.

Reference counts/TPMs are based entirely on reference GTF provided by a user.
Discovered transcript counts/TPMs are based on OUT.transcript_models.gtf file, which contains both known and novel transcripts discovered by IsoQuant. By "discovered known" IsoQuant implies "expressed" reference transcripts.

So, there will be an overlap, possibly quite significant.

As to difference in TPM values - it's expected since IsoQuant uses slightly different read assignment process when using the reference annotation (first stage) and constructing transcript models (second stage). Hence the discrepancy. I plan to work on this aspect in the future as it caused some confusion in the past.

In other words, these output files should not be concatenated. I recommend simply sticking to one of them depending on whether you are interested in novel transcripts or not.

Best
Andrey

[andysaiz]

Thank you so much! This is so helpful, I really appreciate your reply. Since I was confused on the reference and novel outputs overlapping (like above), I chose to run IsoQuant in de novo discovery mode without the reference, then I put the output into SQANTI3 for QC and filtering.

Regardless of whether you use IsoQuant in reference-guided mode or de novo discovery mode, is it recommended to still perform QC and filtering (e.g. SQANTI3)? My understanding is that while IsoQuant outputs the OUT.novel_vs_known.SQANTI-like.tsv file, with some classifications, it is not equivalent to running the whole SQANTI3 pipeline, which performs QC, filtering, and rescue.

[andrewprzh]

Yes, you are right, SQANTI-like output is only provided to get some limited information about generated transcripts, but many columns are filled with default values. So yes, IsoQuant does implement SQANTI functionality.