Inconsistent GTF from UCSC browser vs genePredToGtf

Question

[dariober]

There seem to be some inconsistencies between GTF records from UCSC table browser and from files generated with genePredToGtf (from UCSC utilities).

For example, on the table browser I selected the vegaGene table to search for transcript OTTHUMT00000097860, and I get

chr1    hg19_vegaGene   start_codon 865692  865694  0.000000    +   .   gene_id "OTTHUMT00000097860"; transcript_id "OTTHUMT00000097860"; 
chr1    hg19_vegaGene   CDS 865692  865716  0.000000    +   2   gene_id "OTTHUMT00000097860"; transcript_id "OTTHUMT00000097860"; 
chr1    hg19_vegaGene   exon    865692  865716  0.000000    +   .   gene_id "OTTHUMT00000097860"; transcript_id "OTTHUMT00000097860"; 
chr1    hg19_vegaGene   CDS 866419  866469  0.000000    +   1   gene_id "OTTHUMT00000097860"; transcript_id "OTTHUMT00000097860"; 
...

If instead I use genePredToGtf the start_codon record seems to be missing:

curl -o - -O http://hgdownload.soe.ucsc.edu/goldenPath/hg19/database/vegaGene.txt.gz | gunzip -c \
| cut -f 2- \
| genePredToGtf file stdin stdout \
| grep 'OTTHUMT00000097860' \
| grep 'codon'

This return the stop codon but not the start:

chr1    stdin   stop_codon  879531  879533  .   +   0   gene_id "SAMD11"; transcript_id "OTTHUMT00000097860"; exon_number "12"; exon_id "OTTHUMT00000097860.12"; gene_name "SAMD11";

Am I missing something?

Answer 1

[Denise CS]

Transcript OTTHUMT00000097860 is incomplete at the 5' end, CDS start not found (check 'remarks' in this page). In the GTF file from Ensembl that information is available as "cds_start_NF".

[dariober]

Thanks a lot for digging this information out, it makes sense. This seems to suggest that the output of genePredToGtf is more accurate than the web browser's (?)

[Denise CS]

Glad to help and learn too :) Not sure about accurate versus non accurate as I'm not familiar with the UCSC utilities or table browser. But I'd expect what I see in the Ensembl browser is what I get from the FTP, via BioMart, REST API, Perl APIs. The underlying database is the same, the mode of access is different.