RefseqGene vs Ensembl for mm9

[Thias]

Hello,

I would like to ask for your opinion regarding a reference transcriptome for mm9. My task is to search for "novel" transcripts in RNA-seq data aligned to mm9, so I am wondering what would be my best option to use as reference?

The latest Ensembl release for mm9 dates back to October 2011, whereas the RefseqGene track is the UCSC Genome Browser is still frequently updated. From both tracks I created gtf files with the genePredToGtf tool.

Suprisingly for me the ancient Ensembl release was much more comprehensive than its up to date RefseqGene counterpart:

• Ensembl67 vs. RefseqGene
  

  gffcompare v0.10.1 | Command line was:
  #gffcompare -R -A -r [...] -s
  #= Summary for dataset: ../2011-10-12_Ensembl67.mm9.gtf
  # Query mRNAs : 93727 in 36660 loci (80362 multi-exon transcripts)
  # (14094 multi-transcript loci, ~2.6 transcripts per locus)
  # Reference mRNAs : 35709 in 23009 loci (32141 multi-exon)
  # Super-loci w/ reference transcripts: 21631
  #-----------------| Sensitivity | Precision |
  Base level: 95.1 | 70.8 |
  Exon level: 95.5 | 65.9 |
  Intron level: 97.2 | 80.0 |
  Intron chain level: 83.0 | 33.2 |
  Transcript level: 82.7 | 31.5 |
  Locus level: 93.4 | 57.9 |
  
  Matching intron chains: 26683
  Matching transcripts: 29520
  Matching loci: 21495
  
  Missed exons: 2979/220514 ( 1.4%)
  Novel exons: 45609/347137 ( 13.1%)
  Missed introns: 1975/199233 ( 1.0%)
  Novel introns: 15350/241928 ( 6.3%)
  Missed loci: 0/23009 ( 0.0%)
  Novel loci: 12559/36660 ( 34.3%)

• RefseqGene vs. Ensembl67
  

  gffcompare v0.10.1 | Command line was:
  #gffcompare -R -A -r [...] -s
  #= Summary for dataset: ../2017-10-17_Refseq.refGene.mm9.gtf
  # Query mRNAs : 37083 in 24285 loci (33220 multi-exon transcripts)
  # (6744 multi-transcript loci, ~1.5 transcripts per locus)
  # Reference mRNAs : 80524 in 24101 loci (75471 multi-exon)
  # Super-loci w/ reference transcripts: 21631
  #-----------------| Sensitivity | Precision |
  Base level: 78.9 | 92.8 |
  Exon level: 70.8 | 93.7 |
  Intron level: 83.5 | 95.4 |
  Intron chain level: 35.4 | 80.3 |
  Transcript level: 36.6 | 79.5 |
  Locus level: 88.1 | 88.4 |
  
  Matching intron chains: 26683
  Matching transcripts: 29499
  Matching loci: 21226
  
  Missed exons: 21443/323510 ( 6.6%)
  Novel exons: 7342/224839 ( 3.3%)
  Missed introns: 5863/231321 ( 2.5%)
  Novel introns: 4799/202335 ( 2.4%)
  Missed loci: 0/24101 ( 0.0%)
  Novel loci: 1276/24285 ( 5.3%)

This difference has had a great impact on the number of transcripts (in total and also "novel" ones), which I am amble to assemble based on the same alignment with StringTie or Cufflinks using the respective gtf files as reference.

What would be your approach/thoughts to/on this? Any help/suggestions/ideas welcome.

Best
Matthias

[dpryan]

Is there a reason you're using mm9 still?

It's fairly unsurprising that old Ensembl releases contain more transcripts than current refseq releases. The former is more comprehensive by design, with refseq typically only containing very well annotated isoforms.

[Thias]

Is there a reason you're using mm9 still?

Tons of own, collaborators' and third-party data still aligned to mm9.

It's fairly unsurprising that old Ensembl releases contain more transcripts than current refseq releases. The former is more comprehensive by design, with refseq typically only containing very well annotated isoforms.

So I conclude, that an Ensembl. vs. RefseqGene comparison of the mm10 would approximately have the same result? I didn't know that RefseqGene is particularly restrictive by design...then Ensembl is probably the better reference to find really unannotated transcripts at novel loci.

Thanks for your help