Cuffcompare fails to properly match reference to itself
I have stumbled upon something that I consider really strange.
I ran cuffcompare (2.2.0) on the Ensembl mouse reference annotation from igenomes (NCBI37, May 14 22:13), using both the same genes.gtf as reference and query (nevermind why; just for some troubleshooting).
What I observed was this:
Now, why are there 180 transcripts in the reference that do not completely match themselves in the same reference?
I find this really strange. And since I have been getting some strange results in other parts of my analysis (regarding fpkm distributions of "=" and "c" transcripts), I would really like to know the reason that the above situation occurs.
I decided to take a closer look at some of the "c" transcripts of the "Ensembl-Ensembl" comparison. Gm8292 is one of the problamatic genes. This gene has two different transcripts registered in the genes.gtf file:
The corresponding cuffcompare annotation in the refmap file for this gene:
So it appears that ENSMUST00000135190 completely matches one of the two annotated transcripts for this gene ("="), while the other (ENSMUST00000162574) is only considered a partial match, even though it obviusly is present in the genes.gtf file.
(One might note that this particular problematic transcript is a transcribed processed pseudogene, but there are also regular protein coding transcripts among the 180 "c" matches.)
Maybe I have misunderstood something fundamentally about how cuffcompare works. One explanation could simply be that every isoform except for the "largest" isoform of a given gene is considered a partial match to that larger isoform. But if that was the case, shouldn't there be more than 180 such alternative isoforms in the mouse genome instead? This makes me think that this is a bug in cuffcompare.
Can anyone shed any light on this? Has anyone seen anything similar? Any theories about what might be going on here?
I ran cuffcompare (2.2.0) on the Ensembl mouse reference annotation from igenomes (NCBI37, May 14 22:13), using both the same genes.gtf as reference and query (nevermind why; just for some troubleshooting).
Code:
cuffcompare -r igenomes/Mus_musculus/Ensembl/NCBIM37/Annotation/Genes/genes.gtf igenomes/Mus_musculus/Ensembl/NCBIM37/Annotation/Genes/genes.gtf
Code:
awk '{print $3}' Ensembl_Ensembl.genes.gtf.refmap | sort | uniq -c
93459 =
180 c
1 class_code
I find this really strange. And since I have been getting some strange results in other parts of my analysis (regarding fpkm distributions of "=" and "c" transcripts), I would really like to know the reason that the above situation occurs.
I decided to take a closer look at some of the "c" transcripts of the "Ensembl-Ensembl" comparison. Gm8292 is one of the problamatic genes. This gene has two different transcripts registered in the genes.gtf file:
Code:
grep "Gm8292" genes.gtf 1 pseudogene exon 55392211 55392351 . + . exon_number "1"; gene_biotype "pseudogene"; gene_id "ENSMUSG00000089944"; gene_name "Gm8292"; transcript_id "ENSMUST00000135190"; transcript_name "Gm8292-201"; tss_id "TSS72358"; 1 transcribed_processed_pseudogene exon 55392245 55392640 . + . exon_number "1"; gene_biotype "pseudogene"; gene_id "ENSMUSG00000089944"; gene_name "Gm8292"; transcript_id "ENSMUST00000162574"; transcript_name "Gm8292-001"; tss_id "TSS50335"; 1 pseudogene exon 55392353 55392727 . + . exon_number "2"; gene_biotype "pseudogene"; gene_id "ENSMUSG00000089944"; gene_name "Gm8292"; transcript_id "ENSMUST00000135190"; transcript_name "Gm8292-201"; tss_id "TSS72358";
Code:
grep "Gm8292" Ensembl_Ensembl.genes.gtf.refmap Gm8292 ENSMUST00000135190 = Gm8292|ENSMUST00000135190 Gm8292 ENSMUST00000135190 c Gm8292|ENSMUST00000162574
(One might note that this particular problematic transcript is a transcribed processed pseudogene, but there are also regular protein coding transcripts among the 180 "c" matches.)
Maybe I have misunderstood something fundamentally about how cuffcompare works. One explanation could simply be that every isoform except for the "largest" isoform of a given gene is considered a partial match to that larger isoform. But if that was the case, shouldn't there be more than 180 such alternative isoforms in the mouse genome instead? This makes me think that this is a bug in cuffcompare.
Can anyone shed any light on this? Has anyone seen anything similar? Any theories about what might be going on here?
