I have merged transcriptome assembly of 728 accessions and compare it with Arabidopsis refernce annotation file (ftp://ftp.ensemblgenomes.org/pub/release-43/plants/gtf/arabidopsis_thaliana) using gffcompare. Its stat' output is:

gffcompare -r /media/waqas/second/Analysis/TAIR10_genome/ensemble_gtf/Arabidopsis_thaliana.TAIR10.42.gtf -o gffcompare stringtie_merged.gtf

Summary for dataset: stringtie_merged.gtf

Query mRNAs : 126199 in 32235 loci (112943 multi-exon transcripts)

(17283 multi-transcript loci, ~3.9 transcripts per locus)

Reference mRNAs : 53516 in 32046 loci (42815 multi-exon)

Super-loci w/ reference transcripts: 27093

              | Sensitivity | Precision  |

    Base level:   100.0     |    86.5    |

    Exon level:   100.0     |    66.0    |

  Intron level:   100.0     |    70.0    |

Intron chain level:    99.6     |    37.8    |

Transcript level:    99.6     |    42.2    |

   Locus level:   100.0     |    89.3    |


 Matching intron chains:   42644
   Matching transcripts:   53316
          Matching loci:   32046

      Missed exons:       0/192402  (  0.0%)
       Novel exons:   27127/342721  (  7.9%)
    Missed introns:       1/132525  (  0.0%)
     Novel introns:   22825/189217  ( 12.1%)
       Missed loci:       0/32046   (  0.0%)
        Novel loci:    3349/32235   ( 10.4%)

 Total union super-loci across all input datasets: 32235 
126199 out of 126199 consensus transcripts written in gffcompare.annotated.gtf (0 discarded as redundant)

Besides that It generated five more files, I have checked the literature and found that class code 'j' reflects novel isoforms but the occurrence of j varies in gffcompare output files:

gffcompare.annotated.gtf (count of j: 1000)

gffcompare.stringtie_merged.gtf.tmap (count of j: 2807)

gffcompare.tracking (count of j: 2876)

How can I get the idea of actual number of novel isoforms?