Is SQANTI3 v5.5.1 QC performing filtering during QC step? 192k vs 306k isoforms discrepancy

[FAFUshiyan]

I am using SQANTI3 v5.5.1 for quality control of long-read transcript models (GTF from IsoQuant). When I inspect the QC log, I see this line:

[INFO:2025-12-15 02:17:32,904] After RTS classification: 192164

However, my input GTF contains ~306,000 transcripts, and the QC report shows ~192k isoforms in the classification/report outputs.

According to the documentation, the QC module (sqanti3_qc.py) is supposed to compute descriptors and structural categories for each transcript (e.g., splice junction support, intrapriming, RT-switching, TSS/TTS features) and produce a classification.txt with all these features. The filtering of transcripts into curated/high-confidence sets is normally handled by the filter module (sqanti3_filter.py), with either rules or machine learning.

SQANTI3.qc_params.txt
Version 5.5.1
Input /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/MMM.qc.illumina_sr_corrected.gtf
Annotation /public/home/ac2yrlbilq/ref_data/Human/RNA-seq/gencode.v48.annotation.gtf
Genome /public/home/ac2yrlbilq/ref_data/Human/GCA_000001405.15_GRCh38_no_alt_analysis_set.fna
MinRefLength 0
ForceIdIgnore True
Aligner minimap2
FLCount NA
Expression /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/NGS_data/star/kallisto/kallisto_out/kallisto_exp
Junction /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/NGS_data/star/SJ/.out.tab
CAGEPeak /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/ref_data/human.refTSS_v3.1.hg38.bed
PolyAMotif /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/ref_data/polyA.list.txt
PolyAPeak /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/atlas.clusters.2.0.GRCh38.96.mapped.bed
IsFusion False
PhyloP /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/hg38.phyloP100way.bed
SkipORF True
ORFInput NA
FASTAused False
Expression /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/NGS_data/star/kallisto/kallisto_out/kallisto_exp
GMAPindex NA
OutputPrefix MMM_muscle_sqanti3_v2
OutputDirectory /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/SQANTI3_QC_out
Coverage /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/NGS_data/star/SJ/.out.tab
CanonicalSites ATAC,GCAG,GTAG
PostTTSWindow 20
GeneName False
ReportType both
RunIsoAnnotLite False
isoAnnotGFF3 NA
ShortReads NA
ShortReadsBAMs /public/home/ac2yrlbilq/project/MMM/SQANTI3_v2/NGS_data/star/SR_bam.fofn
ratioTSSmetric max

[pabloati]

Hi, SQANTI3 might eliminate isoforms if they do not have strand information in the input GTF, as they cannot be compared against the reference.

However, could you please repeat your analysis on the latest version of SQANTI (currently v5.5.4)? That way I can give you a better support in case it is due to a bug.

Best,
Pablo

[FAFUshiyan]

OK, thank you.