Structure is as GFF,
so the fields are:
<seqname> <source> <feature> <start> <end> <score>
<strand> <frame> [attributes] [comments]
Here is a simple example with 3 translated exons. Order of rows is not important.
381 Twinscan CDS 380 401 . + 0 gene_id "001"; transcript_id "001.1"; 381 Twinscan CDS 501 650 . + 2 gene_id "001"; transcript_id "001.1"; 381 Twinscan CDS 700 707 . + 2 gene_id "001"; transcript_id "001.1"; 381 Twinscan start_codon 380 382 . + 0 gene_id "001"; transcript_id "001.1"; 381 Twinscan stop_codon 708 710 . + 0 gene_id "001"; transcript_id "001.1";The whitespace in this example is provided only for readability. In GTF, fields must be separated by a single TAB and no white space.
<seqname>
The name of the sequence. Commonly, this is the chromosome ID or contig ID. Note that the coordinates used must be unique within each sequence name in all GTFs for an annotation set.
<source>
The source column should be a unique label indicating where the annotations
came from --- typically the name of either a prediction program or a public
database.
<feature>
The following feature types are required: "CDS", "start_codon", "stop_codon". The features "5UTR", "3UTR", and "exon" are optional. All other features will be ignored.
CDS represents the coding sequence starting with the first translated codon and proceeding to the last translated codon. Unlike Genbank annotation, the stop codon is not included in the CDS for the terminal exon. The optional features "5UTR" and "3UTR" represent exons in the 5' and 3' untranslated region, respectively.
The feature "exon" more generically describes any transcribed exon. Therefore, exon boundaries will be the transcription start site, splice donor, splice acceptor and poly-adenylation site. The start or stop codon will not necessarily lie on an exon boundary.
The "start_codon" feature is up to 3bp long in total and is included in the coordinates for the "CDS" features. The "stop_codon" feature similarly is up to 3bp long and is included in the coordinates for the "3UTR" features, if used.
The "start_codon" and "stop_codon" features are not required to be atomic; they may be interrupted by valid splice sites. A split start or stop codon appears as two distinct features.
<start> <end>
Integer start and end coordinates of the feature relative to the beginning
of the sequence named in <seqname>. <start> must be less than
or equal to <end>. Sequence numbering starts at 1. Values of <start>
and <end> that extend outside the reference sequence are technically
acceptable, but they are discouraged.
<score>
The score field indicates a degree of confidence in the feature's existence and coordinates. The value of this field has no global togel sgp scale but may have relative significance when the <source> field indicates the prediction program used to create this annotation. It may be a floating point number or integer, and not necessary and may be replaced with a dot.
<frame>
0 indicates that the feature begins with a whole codon at the 5' most base.
1 means that there is one extra base (the third base of a codon)
before the first whole codon
and 2 means that there are two extra bases (the second and third bases of the codon) before the first codon. Note that for reverse strand features, the 5' most base is the <end> coordinate.
Here are the details excised from the GFF spec. Important: Note comment on reverse strand.
'0' indicates that the specified region is in frame, i.e. that its first base corresponds to the first base of a codon. '1' indicates that there is one extra base, i.e. that the second base of the region corresponds to the first base of a codon, and '2' means that the third base of the region is the first base of a codon. If the strand is '-', then the first base of the region is value of <end>, because the corresponding coding region will run from <end> to <start> on the reverse strand.
Frame is calculated as (3 - ((length-frame) mod 3)) mod 3.
Attributes must end in a semicolon which must then be separated from the start of any subsequent attribute by exactly one space character (NOT a tab character).
Textual attributes should be surrounded by doublequotes. [comments] Comments begin with a hash ('#') and continue to the end of the line. Nothing beyond a hash will be parsed. These may occur anywhere in the file, including at the end of a feature line.
Here is an example of a gene on the negative strand including UTR regions. Larger coordinates are 5' of smaller coordinates. Thus, the start codon is 3 bp with largest coordinates among all those bp that fall within the CDS regions. Similarly, the stop codon is the 3 bp with coordinates just less than the smallest coordinates within the CDS regions and the largest coordenates among all those within the 3UTR regions.
140 Twinscan 3UTR 65149 65487 . - . gene_id "140.000"; transcript_id "140.000.1";
140 Twinscan 3UTR 66823 66995 . - . gene_id "140.000"; transcript_id "140.000.1";
140 Twinscan stop_codon 66993 66995 . - . gene_id "140.000"; transcript_id "140.000.1";
140 Twinscan CDS 66996 66999 . - 1 gene_id "140.000"; transcript_id "140.000.1";
140 Twinscan CDS 70207 70294 . - 2 gene_id "140.000"; transcript_id "140.000.1";
140 Twinscan CDS 71696 71807 . - 0 gene_id "140.000"; transcript_id "140.000.1";
140 Twinscan start_codon 71805 71807 . - . gene_id "140.000"; transcript_id "140.000.1";
140 Twinscan 5UTR 71808 71812 . - . gene_id "140.000"; transcript_id "140.000.1";
Note the frames of the coding exons. For example:
381 Twinscan exon
150 200 . + . gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan exon
300 401 . + . gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan CDS 380 401 . + 0 gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan exon
501 650 . + . gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan CDS
501 650 . + 2 gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan exon
700 800 . + . gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan CDS
700 707 . + 2 gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan exon
900 1000 . + . gene_id
"381.000"; transcript_id "381.000.1";
381 Twinscan start_codon 380 382
. + 0 gene_id "381.000"; transcript_id
"381.000.1";
381 Twinscan stop_codon 708
710 . + 0 gene_id "381.000";
transcript_id "381.000.1";
Several Perl scripts have been written for checking, parsing, correcting, and comparing GTF-formatted annotations. Most of the important ones are included the Eval package, which comes equipped with a GTF parsing Perl package GTF.pm.
The Eval documentation contains a complete code-level documentation of GTF.pm, suitable for able Perl programmers to create and parse GTF files.