Parsing GFF Files · Biopython

Note: GFF parsing is not yet integrated into Biopython. This documentation is work towards making it ready for inclusion. You can retrieve the current version of the GFF parser from: http://github.com/chapmanb/bcbb/tree/master/gff, which in turn led to https://github.com/daler/gffutils. Comments are very welcome.

Generic Feature Format (GFF) is a biological sequence file format for representing features and annotations on sequences. It is a tab delimited format, making it accessible to biologists and editable in text editors and spreadsheet programs. It is also well defined and can be parsed via automated programs. GFF files are available from many of the large sequencing and annotation centers. The specification provides full details on the format and its uses.

Biopython provides a full featured GFF parser which will handle several versions of GFF: GFF3, GFF2, and GTF. It supports writing GFF3, the latest version.

GFF parsing differs from parsing other file formats like GenBank or PDB in that it is not record oriented. In a GenBank file, sequences are broken into discrete parts which can be parsed as a whole. In contrast, GFF is a line oriented format with support for nesting features. GFF is also commonly used to store only biological features, and not the primary sequence.

These differences have some consequences in how you will deal with GFF:

The documentation below provides a practical guide to examining, parsing and writing GFF files in Python.

Since GFF is a very general format, it is extremely useful to start by getting a sense of the type of data in the file and how it is structured. GFFExaminer provides an interface to examine and query the file. To examine relationships between features, examine a dictionary mapping parent to child features:

This file contains a flexible three level description of coding sequences: genes have mRNA trasncripts; those mRNA transcripts each contain common features of coding sequence, the CDS itself, exon, intron and 5’ and 3’ untranslated regions. This is a common GFF structure allowing representation of multiple transcripts:

Another item of interest for designing your parse strategy is understanding the various tags used to label the features. These consist of:

The available_limits function in the examiner gives you a high level summary of these feature attributes, along with counts for the number of times they appear in the file:

Generally, the GFF parser works similar to other parsers in Biopython. Calling parse with a handle to a GFF file returns a set of SeqRecord objects corresponding to the various IDs referenced in the file:

The rec object is a Biopython SeqRecord containing the features described in the GFF file. The features are ordered into parent-child relationships based on the line by line information in the original GFF file. See the detailed documentation on SeqRecord and SeqFeature objects for more details on accessing the information in these objects.

Since a GFF file is not broken down into an explicit record structure, this requires reading the entire file, parsing all of the features, and then returning those as records. This will be fine for small files, but for most real life cases you will want to restrict parsing to a set of features of interest or a section of lines at once to conserve memory.

A GFF file will commonly contain many types of features, and you will be interested in retrieving a subset of these. The limit_info argument to GFF.parse allows exact specification of which features to parse, turn into objects and retrieve. An example is retrieving all coding sequence on chromosome 1:

You will get back a single record for chromosome 1 which contains all of the coding features in memory for further manipulation. Depending on your memory requirements and workflow, it may make sense to do analysis over each chromosome or set of features you are interested in.

Another way to break up a large GFF file parse into sections is to limit the number of lines that are read at once. This is a useful workflow for GFF files in which you don’t need all of the features at once and can do something useful with a few at a time. To do this, pass the target_lines argument to GFF.parse:

The parser will attempt to smartly break up the file at your requested number of lines. For instance, if 1000 lines happens to come in the middle of a nested coding feature (gene -> transcript -> CDS/exon/intron), the parser would continue until the entire feature region is read. This helps ensure that you have fully formed features for analysis.

If your file has no nesting of features, or you just want a single line at once, you can set target_lines=1 and the parser will happily give you back a SeqRecord object with a single SeqFeature for every line.

GFF records normally contain annotation data, while sequence information is available in a separate FASTA formatted file. The GFF parser can add annotations to existing records. First parse the sequence file with SeqIO, then feed the resulting sequence dictionary to the GFF parser:

Note that this just adds directly to the existing dictionary. If you apply filters to the GFF parser these are only applied to annotations; records will not be removed from the initial sequence dictionary.

The GFF3Writer takes an iterator of SeqRecord objects, and writes each SeqFeature as a GFF3 line:

The remaining qualifiers are the final key/value pairs of the attribute.

A feature hierarchy is represented as sub_features of the parent feature. This handles any arbitrarily deep nesting of parent and child features.

You can create Biopython SeqRecord and SeqFeature objects from scratch and use these to generate GFF output. Chapter 4 of the Tutorial goes into detail about the objects, and this example demonstrates the major features:

This generates the following GFF: