Rsubread featureCounts with built-in mm10 vs Ensemble GRCm38.96.gff3

Question

[akh22]

I am trying to get counts from Bam files generated by STAR. I used Rsubread's featureCount with Ensemble Mus_musculus.GRCm38.96.gff3 ;

fc<-featureCounts(files = "10-IT-1-21-19.bam", nthreads = 24, isPairedEnd = T, isGTFAnnotationFile=T, annot.ext="Mus_musculus.GRCm38.96.gff3", GTF.attrType = "Name"))

and here is a stat;

                 Status Skin.25.IT.1.21.19.bam 
1                       Assigned                7686415           
2            Unassigned_Unmapped                      0                 
3      Unassigned_MappingQuality                      0               
4             Unassigned_Chimera                      0                   
5      Unassigned_FragmentLength                      0                 
6           Unassigned_Duplicate                      0                    
7        Unassigned_MultiMapping                      0                     
8           Unassigned_Secondary                      0                    
9            Unassigned_NonSplit                      0                    
10         Unassigned_NoFeatures                2985131

If I run this with buil-in mm10 as follows;

fc<-featureCounts(files = "10-IT-1-21-19.bam", nthreads = 24, isPairedEnd = T, annot.inbuilt = "mm10"))

and stat for this is;

Status Skin.25.IT.1.21.19.bam
1                       Assigned               18096135
2            Unassigned_Unmapped                      0
3      Unassigned_MappingQuality                      0
4             Unassigned_Chimera                      0
5      Unassigned_FragmentLength                      0
6           Unassigned_Duplicate                      0
7        Unassigned_MultiMapping                      0
8           Unassigned_Secondary                      0
9            Unassigned_NonSplit                      0
10         Unassigned_NoFeatures                4960586
11 Unassigned_Overlapping_Length                      0
12          Unassigned_Ambiguity                 911357

As you can see # of assigned by Ensemble Mus_musculus.GRCm38.96.gff3 is significantly less than the one by mm10. I thought Ensemble gff3 annotation has a larger coverage than mm10. I'd appreciate any comments on this.

Thanks.

[swbarnes2]

What reference was used to make the bam?

[akh22]

I believe it was ensemble GRcm38, though I am not 100% certain since the read mapping was done by some other lab.

[Devon Ryan]

Try the GTF file from Ensembl instead (never use GFF files unless you have no other choice).

[akh22]

What is a reason for using GTF over GFF3 ? I thought GFF3 was a preferred annotation file over GTF.

Thanks.

[akh22]

So I tried the ensemble GTF and got on the average 80% assigned reads. GTF looks definitely promising than GFF3.

[Devon Ryan]

It's kind of a dirty little secret that GFF files are tough to support.

[akh22]

Yeah, I am begging to realize this, despite of what I was told from "experts".

Thanks.