Transcript classification codes ? about gffcompare HOT 1 CLOSED

If you are looking for novel alternative splicing isoform (with novel junctions), code j might be the more "promissing" one. See the table at http://ccb.jhu.edu/software/stringtie/gffcompare.shtml#transfrag-class-codes
However there are other codes that can potentially capture that - even k (reverse containment) or o (other unclassified overlap) could be revealed to be a novel isoform. Of course, transcriptional noise, alignment artifacts and even assembly issues (fragmentation, coverage bias etc.) should be considered.

One could argue that even code = (matching) could in fact hide a potentially novel isoform - one with a different TSS and/or TES, perhaps.

So the most disappointingly encompassing answer to your question could be: "all of them".

from gffcompare.