Introduction

gffutils is a Python package for working with GFF and GTF files in a hierarchical manner. It allows operations which would be complicated or time-consuming using a text-file-only approach.

Below is a short demonstration of gffutils. For the full documentation, see the rest of the table of contents to the left.

Example file

Consider the first gene on chromosome 2L in Drosophila melanogaster, CG11023. Here is a graphical representation of this gene’s three alternative transcripts, from FlyBase :

_images/flybase-FBgn0031208.png

gffutils ships with the GFF annotation of this gene, downloaded from release 5.54 of FlyBase. We’ll be using it to describe some features of gffutils, because it contains just enough complexity to be representative of a real-world, genome-wide GFF file.

Let’s look at what the file contains:

>>> import gffutils
>>> fn = gffutils.example_filename('intro_docs_example.gff')
>>> print(open(fn).read()) 
2L      FlyBase gene    7529    9484    .       +       .       ID=FBgn0031208;Name=CG11023;Ontology_term=SO:0000010,SO:0000087,GO:0016929,GO:0016926,GO:0006508;Dbxref=FlyBase:FBan0011023,FlyBase_Annotation_IDs:CG11023,GB_protein:ACZ94128,GB_protein:AAO41164,GB:AI944728,GB:AJ564667,GB_protein:CAD92822,GB:BF495604,UniProt/TrEMBL:Q86BM6,INTERPRO:IPR003653,GB_protein:AGB92323,UniProt/TrEMBL:M9PAY1,OrthoDB7_Drosophila:EOG796K1P,OrthoDB7_Diptera:EOG7X1604,EntrezGene:33155,UniProt/TrEMBL:E1JHP8,UniProt/TrEMBL:Q6KEV3,OrthoDB7_Insecta:EOG7Q8QM7,OrthoDB7_Arthropoda:EOG7R5K68,OrthoDB7_Metazoa:EOG7D59MP,InterologFinder:33155,BIOGRID:59420,FlyAtlas:CG11023-RA,GenomeRNAi:33155;gbunit=AE014134;derived_computed_cyto=21A5-21A5
2L      FlyBase mRNA    7529    9484    .       +       .       ID=FBtr0300689;Name=CG11023-RB;Parent=FBgn0031208;Dbxref=REFSEQ:NM_001169365,FlyBase_Annotation_IDs:CG11023-RB;score_text=Strongly Supported;score=11
2L      FlyBase mRNA    7529    9484    .       +       .       ID=FBtr0300690;Name=CG11023-RC;Parent=FBgn0031208;Dbxref=REFSEQ:NM_175941,FlyBase_Annotation_IDs:CG11023-RC;score_text=Strongly Supported;score=15
2L      FlyBase mRNA    7529    9484    .       +       .       ID=FBtr0330654;Name=CG11023-RD;Parent=FBgn0031208;Dbxref=FlyBase_Annotation_IDs:CG11023-RD,REFSEQ:NM_001272857;score_text=Strongly Supported;score=11
2L      FlyBase exon    7529    8116    .       +       .       Name=CG11023:1;Parent=FBtr0300689,FBtr0300690,FBtr0330654;parent_type=mRNA
2L      FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0300689,FBtr0300690;parent_type=mRNA
2L      FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase CDS     7680    8116    .       +       0       Name=CG11023-cds;Parent=FBtr0300689,FBtr0300690;parent_type=mRNA
2L      FlyBase CDS     7680    8116    .       +       0       Name=CG11023-cds;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase intron  8117    8228    .       +       .       Name=CG11023-in;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase intron  8117    8192    .       +       .       Name=CG11023-in;Parent=FBtr0300689;parent_type=mRNA
2L      FlyBase intron  8117    8192    .       +       .       Name=CG11023-in;Parent=FBtr0300690;parent_type=mRNA
2L      FlyBase exon    8193    9484    .       +       .       Name=CG11023:3;Parent=FBtr0300689;parent_type=mRNA
2L      FlyBase CDS     8193    8610    .       +       1       Name=CG11023-cds;Parent=FBtr0300689;parent_type=mRNA
2L      FlyBase CDS     8193    8589    .       +       1       Name=CG11023-cds;Parent=FBtr0300690;parent_type=mRNA
2L      FlyBase exon    8193    8589    .       +       .       Name=CG11023:2;Parent=FBtr0300690;parent_type=mRNA
2L      FlyBase exon    8229    9484    .       +       .       Name=CG11023:4;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase CDS     8229    8610    .       +       1       Name=CG11023-cds;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase intron  8590    8667    .       +       .       Name=CG11023-in;Parent=FBtr0300690;parent_type=mRNA
2L      FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300689;parent_type=mRNA
2L      FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase three_prime_UTR 9277    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300690;parent_type=mRNA
2L      FlyBase exon    8668    9484    .       +       .       Name=CG11023:5;Parent=FBtr0300690;parent_type=mRNA
2L      FlyBase CDS     8668    9276    .       +       0       Name=CG11023-cds;Parent=FBtr0300690;parent_type=mRNA

While this is a (relatively) simple gene, the annotation looks quite complex. In no particular order, some comments about this annotation:

  • The gene has lots of attributes (last column), which would be nice to be able to access from Python.

  • The gene and mRNA features have “ID” attributes, while UTRs, exons, CDSs, and introns only have “Name” attributes, e.g.:

    gene: ID=FBgn0031208;Name=CG11023...
    mRNA: ID=FBtr0300689;Name=CG11023-RB...
    exon: Name=CG11023:1...
    
  • The “Name” attributes of UTRs, exons, CDSs, and introns contain the “Name” attribute of the parent gene, followed by some extra info. For exons, it’s a colon and a number; for introns its “-in”, and so on. But the “Parent” mRNA is indicated with the mRNA’s “ID” attribute, e.g.,:

    exon: Name=CG11023:3;Parent=FBtr0300689;
    

    For routine tasks like counting reads in exons (say, with HTSeq), it would be nice to be able to add a “gene_id” attribute to exons. That is, change:

    exon: Name=CG11023:3;Parent=FBtr0300689;
    

    to:

    exon: Name=CG11023:3;Parent=FBtr0300689;gene_id=FBgn0031208;
    

    Then, we could use htseq-count like this:

    htseq-count --type=exon --idattr=gene_id reads.sam fixed_annotation.gff
    
  • UTRs, exons, CDSs, and introns have the same name for different features. For example, these two introns have different coordinates and have different mRNA parents, yet have the same “Name” attribute:

    2L  FlyBase intron  8117    8228    .       +       .       Name=CG11023-in;Parent=FBtr0330654;parent_type=mRNA
    2L  FlyBase intron  8117    8192    .       +       .       Name=CG11023-in;Parent=FBtr0300689;parent_type=mRNA
    

    It would be nice to have unique IDs for these features so we could differentiate them for, say, counting the unique introns of a gene or generating a file of splice junctions.

Create the database

The first step to working with gffutils is to import the file into a local sqlite3 file-based database. The gffutils.create_db() function is used for this, and it takes many optional arguments for configuring how to interpret your GFF or GTF file.

>>> db = gffutils.create_db(fn, dbfn='test.db', force=True, keep_order=True,
... merge_strategy='merge', sort_attribute_values=True)

Here, force=True will overwrite any existing database, keep_order=True will maintain the order of the attributes (ID, Name, Parent, and so on), and sort_attribute_values=True will ensure the values of each attribute are always sorted. These latter two settings are necessary here for this documentation (which undergoes automated testing) to have predictable output, but in general you can gain a speedup by using the default keep_order=False and sort_attributes_order=False.

If your input file follows the GFF or GTF file format specifications, this is all you need to create a database. But real-world files don’t always completely adhere to the format specifications. So gffutils provides lots of configuration options to tailor the pre-processing to your particular file – see Importing data into a database for more details, and Examples for more practical examples.

Use the database

Once the database has been created, downstream code can connect to it simply by:

>>> import gffutils
>>> db = gffutils.FeatureDB('test.db', keep_order=True)

Features are accessed by name:

>>> gene = db['FBgn0031208']
>>> gene
<Feature gene (2L:7529-9484[+]) at 0x...>

Inspect fields from the GFF line:

>>> gene.start
7529
>>> gene.end
9484

Attributes are stored in the attributes dictionary. Values are always in a list, even if there’s only one item:

>>> gene.attributes['Name']
['CG11023']

Attributes can also be accessed from the Feature object itself, to save some typing:

>>> gene['Name']
['CG11023']

Printing a Feature reproduces the original GFF line as faithfully as possible. See the Dialects section for more details on how this is handled and configured.

>>> print(gene)  
2L      FlyBase gene    7529    9484    .       +       .       ID=FBgn0031208;Name=CG11023;Ontology_term=SO:0000010,SO:0000087,GO:0016929,GO:0016926,GO:0006508;Dbxref=FlyBase:FBan0011023,FlyBase_Annotation_IDs:CG11023,GB_protein:ACZ94128,GB_protein:AAO41164,GB:AI944728,GB:AJ564667,GB_protein:CAD92822,GB:BF495604,UniProt/TrEMBL:Q86BM6,INTERPRO:IPR003653,GB_protein:AGB92323,UniProt/TrEMBL:M9PAY1,OrthoDB7_Drosophila:EOG796K1P,OrthoDB7_Diptera:EOG7X1604,EntrezGene:33155,UniProt/TrEMBL:E1JHP8,UniProt/TrEMBL:Q6KEV3,OrthoDB7_Insecta:EOG7Q8QM7,OrthoDB7_Arthropoda:EOG7R5K68,OrthoDB7_Metazoa:EOG7D59MP,InterologFinder:33155,BIOGRID:59420,FlyAtlas:CG11023-RA,GenomeRNAi:33155;gbunit=AE014134;derived_computed_cyto=21A5-21A5

Get all the mRNAs for a gene:

>>> for i in db.children(gene, featuretype='mRNA', order_by='start'):
...     print(i) 
2L      FlyBase mRNA    7529    9484    .       +       .       ID=FBtr0300689;Name=CG11023-RB;Dbxref=REFSEQ:NM_001169365,FlyBase_Annotation_IDs:CG11023-RB;Parent=FBgn0031208;score_text=Strongly Supported;score=11
2L      FlyBase mRNA    7529    9484    .       +       .       ID=FBtr0300690;Name=CG11023-RC;Dbxref=REFSEQ:NM_175941,FlyBase_Annotation_IDs:CG11023-RC;Parent=FBgn0031208;score_text=Strongly Supported;score=15
2L      FlyBase mRNA    7529    9484    .       +       .       ID=FBtr0330654;Name=CG11023-RD;Dbxref=FlyBase_Annotation_IDs:CG11023-RD,REFSEQ:NM_001272857;Parent=FBgn0031208;score_text=Strongly Supported;score=11

Get all the exons for a gene:

>>> for i in db.children(gene, featuretype='exon', order_by='start'):
...     print(i) 
2L      FlyBase exon    7529    8116    .       +       .       Name=CG11023:1;Parent=FBtr0300689,FBtr0300690,FBtr0330654;parent_type=mRNA
2L      FlyBase exon    8193    9484    .       +       .       Name=CG11023:3;Parent=FBtr0300689;parent_type=mRNA
2L      FlyBase exon    8193    8589    .       +       .       Name=CG11023:2;Parent=FBtr0300690;parent_type=mRNA
2L      FlyBase exon    8229    9484    .       +       .       Name=CG11023:4;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase exon    8668    9484    .       +       .       Name=CG11023:5;Parent=FBtr0300690;parent_type=mRNA

Does this gene have any constitutive exons (exons found in every alternative transcript)? This is an example of something that would be difficult to do using only standard text file processing tools, and demonstrates the usefulness of the gffutils:

>>> mRNA_count = len(list(db.children(gene, featuretype='mRNA')))
>>> constitutive_exons = []
>>> for exon in db.features_of_type('exon', order_by='start'):
...     parents = db.parents(exon, featuretype='mRNA')
...     if len(list(parents)) == mRNA_count:
...         constitutive_exons.append(exon)
>>> print(constitutive_exons)
[<Feature exon (2L:7529-8116[+]) at 0x...>]

Retrieve entries by genomic coordinates, which uses the UCSC binning strategy:

>>> list(db.region(region=('2L', 9277, 10000), completely_within=True))
[<Feature three_prime_UTR (2L:9277-9484[+]) at 0x...>]

Retrieve UTRs that overlap the gene (note that a feature can be provided as the coordinates):

>>> for UTR in db.region(gene, featuretype=['three_prime_UTR', 'five_prime_UTR']):
...     print(UTR) 
2L      FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0300689,FBtr0300690;parent_type=mRNA
2L      FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300689;parent_type=mRNA
2L      FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase three_prime_UTR 9277    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300690;parent_type=mRNA

Now let’s illustrate a problem in the GFF file as it is currently formatted by trying to get the consitutive 5’UTRs in the same way we got the constitutive exons above:

>>> constitutive_5UTRs = []
>>> for utr in db.features_of_type('five_prime_UTR', order_by='start'):
...     parents = db.parents(utr, featuretype='mRNA')
...     if len(list(parents)) == mRNA_count:
...         constitutive_5UTRs.append(utr)
>>> print(constitutive_5UTRs)
[]

Yikes! What happened? This is because by default, each line in the GFF file is treated as a unique feature. From this perspective, the first 5’UTR has two parents, and the second 5’UTR has one parent – there are zero 5’UTRs with all three transcripts as parents.

What we’d really like to do is to merge the features that have identical IDs and coordinates. Specifically, we’d like to have these two features:

2L  FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0300689,FBtr0300690;parent_type=mRNA
2L  FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0330654;parent_type=mRNA

become a single feature:

2L  FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0300689,FBtr0300690,FBtr0330654;parent_type=mRNA

Similarly, we’d like these three 3’UTRs:

2L  FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300689;parent_type=mRNA
2L  FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0330654;parent_type=mRNA
2L  FlyBase three_prime_UTR 9277    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300690;parent_type=mRNA

to become two (merging only the first two because they have identical coordinates):

2L  FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300689,FBtr0330654;parent_type=mRNA
2L  FlyBase three_prime_UTR 9277    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300690;parent_type=mRNA

In order to do this, we need to change how features are loaded into the database by using merge_strategy=True and id_spec=['ID', 'Name']. You can read more about exactly what these arguments do, and other ways of tweaking the import, at Database IDs. But for now, let’s see what happens:

>>> db2 = gffutils.create_db(fn, dbfn='test.db', force=True, keep_order=True,
... sort_attribute_values=True,
... merge_strategy='merge',
... id_spec=['ID', 'Name'])
>>> for utr in db2.features_of_type('five_prime_UTR', order_by='start'):
...     parents = db2.parents(utr, featuretype='mRNA')
...     if len(list(parents)) == mRNA_count:
...         constitutive_5UTRs.append(utr)
>>> print(constitutive_5UTRs)
[<Feature five_prime_UTR (2L:7529-7679[+]) at 0x...>]

Let’s look at all the 5’UTRs in the database to see how the merge strategy affected them:

>>> for i in db2.features_of_type('five_prime_UTR'):
...
...     print(i) 
2L      FlyBase five_prime_UTR  7529    7679    .       +       .       Name=CG11023-u5;Parent=FBtr0300689,FBtr0300690,FBtr0330654;parent_type=mRNA

Aha! A single 5’UTR. What about 3’UTRs?

>>> for i in db2.features_of_type('three_prime_UTR'):
...     print(i) 
2L      FlyBase three_prime_UTR 8611    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300689,FBtr0330654;parent_type=mRNA
2L      FlyBase three_prime_UTR 9277    9484    .       +       .       Name=CG11023-u3;Parent=FBtr0300690;parent_type=mRNA
>>> for i in db2.features_of_type('intron'):
...     print(i) 
2L      FlyBase intron  8117    8228    .       +       .       Name=CG11023-in;Parent=FBtr0330654;parent_type=mRNA
2L      FlyBase intron  8117    8192    .       +       .       Name=CG11023-in;Parent=FBtr0300689,FBtr0300690;parent_type=mRNA
2L      FlyBase intron  8590    8667    .       +       .       Name=CG11023-in;Parent=FBtr0300690;parent_type=mRNA