Querying Ensembl

Section author: Gavin Huttley, Hua Ying

We begin this documentation with a note on dependencies, performance and code status. Ensembl makes their data available via MySQL servers, so the cogent.db.ensembl module has additional dependencies of mysql-connector-python or MySQL-python and SQLAlchemy. You can use pip to install both mysql-connector-python and SQLAlchemy, but MySQL-python is more involved. If you experience trouble, please create a new issue on the issue tracker and tag it with the question tag. Regarding performance, significant queries to the UK servers from Australia are very slow. The examples in this documentation, for instance, take ~15 minutes to run when pointed at the UK servers. Running against a local installation, however, is ~50 times faster. On status, the cogent.db.ensembl module should be considered beta level code. We still strongly advise users to check results for a subset of their analyses against the data from the UK Ensembl web site.

The major components of Ensembl are compara and individual genomes. In all cases extracting data requires connecting to MySQL databases on a server and the server may be located remotely or locally. For convenience, the critical objects you’ll need to query a database are provided in the top-level module import, ie immediately under cogent.db.ensembl.

Specifying a Host and Account

So the first step is to specify what host and account are to be used. On my lab’s machines, I have set an environment variable with the username and password for our local installation of the Ensembl MySQL databases, e.g. ENSEMBL_ACCOUNT="host username password". So I’ll check for that (since the documentation runs much quicker when this is true) and if it’s absent, we just set account=None and the account used defaults to the UK Ensembl service. I also define which release of ensembl we’ll use in one place to allow easier updating of this documentation.

>>> import os
>>> Release = 81
>>> from cogent.db.ensembl import HostAccount
>>> if 'ENSEMBL_ACCOUNT' in os.environ:
...     host, username, password = os.environ['ENSEMBL_ACCOUNT'].split()
...     account = HostAccount(host, username, password)
... else:
...     account = None

What Species Are Available?

Another key element, of course, is the species available. Included as part of cogent.db.ensembl is the module species. This module contains a class that translates between latin names, common names and ensembl database prefixes. The species listed are guaranteed to be incomplete, given Ensembl’s update schedule, so it’s possible to dynamically add to this listing, or even change the common name for a given latin name.

>>> from cogent.db.ensembl import Species
>>> print Species
       Common Name                   Species Name              Ensembl Db Prefix
         A.aegypti                  Aedes aegypti                  aedes_aegypti
        A.clavatus           Aspergillus clavatus           aspergillus_clavatus...

In Australia, the common name for Gallus gallus is chook, so I’ll modify that.

>>> Species.amendSpecies('Gallus gallus', 'chook')
>>> assert Species.getCommonName('Gallus gallus') == 'chook'

You can also add new species for when they become available using Species.amendSpecies.

Species common names are used to construct attributes on PyCogent Compara instances). You can get the name that will be using the getComparaName method. For species with a real common name

>>> Species.getComparaName('Procavia capensis')

or with a shortened species name

>>> Species.getComparaName('Caenorhabditis remanei')

The Species class is basically used to translate between latin names and ensembl’s database naming scheme. It also serves to allow the user to simply enter the common name for a species in order to reference it’s genome databases. The queries are case-insensitive.

Interrogating a Genome

As implied above, Ensembl databases are versioned, hence you must explicitly state what release you want. Aside from that, getting an object for querying a genome is simply a matter of importing the HostAccount and Genome classes. Here I’m going to use the cogent.db.ensembl level imports.

>>> from cogent.db.ensembl import HostAccount, Genome
>>> human = Genome(Species='human', Release=Release, account=account)
>>> print human
Genome(Species='Homo sapiens'; Release='81')

Notice I used the common name rather than full name. The Genome provides an interface to obtaining different attributes. It’s primary role is to allow selection of genomic regions according to some search criteria. The type of region is presently limited to Gene, Est, CpGisland, Repeat and Variation. There’s also a GenericRegion. The specific types are also capable of identifying information related to themselves, as we will demonstrate below.

A Note on Coordinate Systems

The positions employed on Ensembl’s web-site, and in their MySQL database differ from those used internally by cogent.db.ensembl. In all cases where you are querying cogent.db.ensembl objects directly inputting nucleotide positions you can indicate you are using Ensembl coordinates by setting ensembl_coord=True. If you are explicitly passing in a cogent.db.ensembl region, that argument has no effect.

Selecting Gene’s

The genome can be queried for gene’s in a number of ways. You can search for genes using the Genome.getGeneByStableId method which requires you know the Ensembl stable id.

>>> brca1 = human.getGeneByStableId(StableId='ENSG00000012048')
>>> print brca1.Description
breast cancer 1, early onset...

Alternatively, you can query using the Genome.getGenesMatching method. This method allows querying for gene(s) by the following identifiers: HGNC symbol; Ensembl stable_id; description; or coding type.


When querying by description, you can specify that the exact words in the query must be present in the description by setting the argument like=True. The default is like=False.

In general for such queries, case shouldn’t matter. For instance, find the BRCA2 gene by it’s HGNC symbol.

>>> genes = human.getGenesMatching(Symbol='brca2')

Because there can be multiple hits from a getGenesMatching query, and because we wish to not spend time doing things (like talking to the database) unnecessarily, the result of the query is a python generator. This acts like a series and allows you to iterate over the database hits until you find the one you want and then terminate the record collection.

>>> for gene in genes:
...     if gene.Symbol.lower() == 'brca2':
...         break
>>> brca2 = gene # so we keep track of this reference for later on
>>> print brca2.Symbol
>>> print brca2.Description
breast cancer 2...
>>> print brca2
Gene(Species='Homo sapiens'; BioType='protein_coding'; Description='breast...

This code serves to illustrate a few things. First, the sorts of properties that exist on the object. These can be directly accessed as illustrated above. Secondly, that the argument names to getGenesMatching match the properties.

Gene’s also have a location. The length of a gene is the difference between its start and end location.


Unfortunately all gene coordinates can vary between genome builds. So start, end and length can all differ between Ensembl releases for the same gene.

>>> print brca2.Location
Homo sapiens:chromosome:13:32315473...
>>> print len(brca2)

Each location is directly tied to the parent genome and the coordinate above also shows the coordinates’ type (chromosome in this case), name (13), start, end and strand. The start and end positions are python indices and will differ from the Ensembl indices in that start will be the Ensembl index - 1. This is because python counts from 0, not 1. In querying for regions using a specific set of coordinates, it is possible to put in the Ensembl coordinates (demonstrated below).

Gene has several useful properties, including the ability to directly get their own DNA sequence and their CanonicalTranscript and Transcripts. CanonicalTranscript is the characteristic transcript for a gene, as defined by Ensembl. Transcripts is a tuple attribute containing individual region instances of type Transcript. A Transcript has Exons, Introns, a Cds and, if the BioType is protein coding, a protein sequence. In the following we grab the cannonical transcript from brca2

>>> print brca2.BioType
>>> print brca2.Seq
>>> print brca2.CanonicalTranscript.Cds
>>> print brca2.CanonicalTranscript.ProteinSeq

It is also possible to iterate over a transcript’s exons, over their translated exons, or to obtain their coding DNA sequence. We grab the second transcript for this.

>>> transcript = brca2.Transcripts[0]
>>> for exon in transcript.Exons:
...     print exon, exon.Location
Exon(StableId=ENSE00001184784, Rank=1) Homo sapiens:chromosome:13:...
>>> for exon in transcript.TranslatedExons:
...     print exon, exon.Location
Exon(StableId=ENSE00001484009, Rank=2) Homo sapiens:chromosome:13:...
>>> print transcript.Cds

The Cds sequence includes the stop-codon, if present. The reason for this is there are many annotated transcripts in the Ensembl database the length of whose transcribed exons are not divisible by 3. Hence we leave it to the user to decide how to deal with that, but mention here that determining the number of complete codons is trivial and you can slice the Cds so that it’s length is divisible by 3.

The Exons and TranslatedExons properties are tuples that are evaluated on demand and can be sliced. Each Exon/TranslatedExon is itself a region, with all of the properties of generic regions (like having a Seq attribute). Similar descriptions apply to the Introns property and Intron class. We show just for the canonical transcript.

>>> for intron in brca2.CanonicalTranscript.Introns:
...     print intron
Intron(TranscriptId=ENST00000380152, Rank=1)
Intron(TranscriptId=ENST00000380152, Rank=2)
Intron(TranscriptId=ENST00000380152, Rank=3)...

The Gene region also has convenience methods for examining properties of it’s transcripts, in presenting the Cds lengths and getting the Transcript encoding the longest Cds.

>>> print brca2.getCdsLengths()
[10257, 10257]
>>> longest = brca2.getLongestCdsTranscript()
>>> print longest.Cds

All Regions have a getFeatures method which differs from that on genome only in that the genomic coordinates are automatically entered for you. Regions also have the ability to return their sequence as an annotated cogent sequence. The method on Gene simply queries the parent genome using the gene’s own location as the coordinate for the currently supported region types. We will query brca2 asking for gene features, the end-result will be a cogent sequence that can be used to obtain the CDS, for instance, using the standard cogent annotation capabilities.

>>> annot_brca2 = brca2.getAnnotatedSeq(feature_types='gene')
>>> cds = annot_brca2.getAnnotationsMatching('CDS')[0].getSlice()
>>> print cds

Those are the properties of a Gene, at present, of direct interest to end-users.

There are obviously different types of genes, and the Genome object provides an ability to establish exactly what distinct types are defined in Ensembl.

>>> print human.getDistinct('BioType')
['unitary_pseudogene', 'rRNA', 'lincRNA',...

The genome can be queried for any of these types, for instance we’ll query for rRNA. We’ll get the first few records and then exit.

>>> rRNA_genes = human.getGenesMatching(BioType='rRNA')
>>> count = 0
>>> for gene in rRNA_genes:
...     print gene
...     count += 1
...     if count == 1:
...         break
Gene(Species='Homo sapiens'; BioType='Mt_rRNA'; Description='mitochondrially encoded 12S...

This has the effect of returning any gene whose BioType includes the phrase rRNA. If a gene is not a protein coding gene, as in the current case, then it’s Transcripts will have ProteinSeq==None and TranslatedExons==None, but it will have Exons and a Cds.

>>> transcript = gene.Transcripts[0]
>>> assert transcript.ProteinSeq == None
>>> assert transcript.TranslatedExons == None
>>> assert transcript.Cds != None

Getting ESTs

Ensembl’s otherfeatures database mirrors the structure of the core database and contains EST information. Hence, the Est region inherits directly from Gene (ie has many of the same properties). est is a supported feature_types for the getFeatures method. You can also directly query for an EST using Ensembl’s StableID. Here, however, we’ll just query for Est that map to the brca2 region.

>>> ests = human.getFeatures(feature_types='est', region=brca2)
>>> for est in ests:
...     print est
Est(Species='Homo sapiens'; BioType='protein_coding'; Description='None';...

Getting Variation

Variation regions also have distinctive properties worthy of additional mention. As for genes, there are distinct types stored in Ensembl that may be of interest. Those types can likewise be discovered from the genome,

>>> print human.getDistinct('Effect')
['incomplete_terminal_codon_variant', 'splice_acceptor_variant', 'intergenic_variant'...

and that information can be used to query the genome for all variation of that effect.


What we term effect, Ensembl terms consequence. We use effect because it’s shorter.

We allow the query to be an inexact match by setting like=True. Again we’ll just iterate over the first few.

>>> nsyn_variants = human.getVariation(Effect='missense_variant', limit=1000)
>>> for i, nsyn_variant in enumerate(nsyn_variants):
...     if nsyn_variant.Effect == 'missense_variant' and\
...                          nsyn_variant.AlleleFreqs:
...         break
>>> print nsyn_variant
Variation(Symbol='rs17852242'; Effect='missense_variant'; Alleles='A/T')
>>> print nsyn_variant.AlleleFreqs
allele      freq    population_id
     A    0.8620            12951
     T    0.1380            12951

Variation objects also have other useful properties, such as a location, the number of alleles and the allele frequencies. The length of a Variation instance is the length of it’s longest allele.

>>> assert len(nsyn_variant) == 1
>>> print nsyn_variant.Location
Homo sapiens:chromosome:CHR_HSCHR12_1_CTG1:70987-70988:1
>>> assert nsyn_variant.NumAlleles == 2

Variation objects have FlankingSeq and Seq attributes which in the case of a SNP is a single nucleotide long and should correspond to one of the alleles. In the latter case, this property is a tuple with the 0th entry being the 5’- 300 nucleotides and the 1st entry being the 3’ nucleotides.


The flanking sequence is only returned when the SNPs flank matches reference (according to Ensembl).

>>> print nsyn_variant.FlankingSeq[0]
>>> print nsyn_variant.FlankingSeq[1]
>>> assert str(nsyn_variant.Seq) in nsyn_variant.Alleles, str(nsyn_variant.Seq)

As a standard feature, Variation within a specific interval can also be obtained. Using the brca2 gene region instance created above, we can find all the genetic variants using the Variants property of genome regions. We use this example to also demonstrate the PeptideAlleles and TranslationLocation attributes. PeptideAlleles is the amino-acid variation resulting from the nucleotide variation while TranslationLocation is the position in the translated peptide of the variant. If a variant does not affect protein coding sequence (either it’s not exonic or it’s a synonymous variant) then these properties have the value None. We illustrate their use.

>>> for variant in brca2.Variants:
...     if variant.PeptideAlleles is None:
...         continue
...     print variant.PeptideAlleles, variant.TranslationLocation
P/L 1...


These are Python coordinates, add 1 to get the Ensembl value.

We can also use a slightly more involved query to find all variants within the gene of a specific type. (Of course, you could also simply iterate over the Variants attribute to grab these out too.)

>>> brca2_snps = human.getFeatures(feature_types='variation',
...                      region=brca2)
>>> for snp in brca2_snps:
...     if 'missense_variant' in snp.Effect:
...         break
>>> print snp
Variation(Symbol='rs80358836'; Effect=['non_coding_transcript_exon_variant', 'missense_variant'...
>>> print snp.Location
Homo sapiens:chromosome:13:32316464-32316465:1

Other Region Types

These can be obtained from the genome instance using the genomes getFeatures method. At present, only repeats, CpG islands, variation, EST’s and genes can be obtained through this method. There’s also GenericRegion, which is precisely that.

In Ensembl’s databases, each type of feature may be recorded at multiple coordinate levels. Accordingly, each level is checked to obtain full information of that feature.

>>> chicken = Genome(Species='chook', Release=Release, account=account)
>>> print chicken.FeatureCoordLevels
Gallus gallus
     Type                  Levels
     gene    chromosome, scaffold
   repeat    chromosome, scaffold
      est    chromosome, scaffold
variation              chromosome
      cpg    chromosome, scaffold

Comparative Analyses

The Ensembl compara database is represented by cogent.db.ensembl.compara.Compara. This object provides a means for querying for relationships among genomes and obtaining multiple alignments. For convenience the class is made available through the top-level module for importing (i.e. cogent.db.ensembl.Compara). Instantiating Compara requires, as before, the ensembl release, the series of species of interest and optionally an account (we also use our local account for speed). For the purpose of illustration we’ll use the human, mouse and rat genomes.


Any queries on this instance of compara will only return results for the indicated species. If you want to query about other species, create another instance.

>>> from cogent.db.ensembl import Compara
>>> compara = Compara(['human', 'mouse', 'rat'], account=account,
...                  Release=Release)
>>> print compara
Compara(Species=('Homo sapiens', 'Mus musculus', 'Rattus norvegicus'); Release=81...

The Compara object loads the corresponding Genome‘s and attaches them to itself as named attributes (use Species.getComparaName to find out what the attribute will be). The genome instances are named according to their common name in CamelCase, or Scase. For instance, if we had created a Compara instance with the American pika species included, then that genome would be accessed as compara.AmericanPika. Common names containing a ‘.’ are treated differently. For instance, the common name for Caenorhabditis remanei is C.remanei which becomes compara.Cremanei. We access the human genome in this Compara instance and conduct a gene search.

>>> brca2 = compara.Human.getGeneByStableId(StableId='ENSG00000139618')
>>> print brca2
Gene(Species='Homo sapiens'; BioType='protein_coding'; Description='breast...

We can now use this result to search compara for related genes. We note here that like Genome, Compara has the getDistinct method to assist in identifying appropriate search criteria. What are the distinct types of gene relationships recorded in Ensembl, for instance?

>>> relationships = compara.getDistinct('relationship')
>>> print relationships
[u'gene_split', u'alt_allele', u'ortholog_one2many', u'ortholog_one2one'...

So we use the brca2 instance above and search for orthologs among the human, mouse, rat genomes.

>>> orthologs = compara.getRelatedGenes(gene_region=brca2,
...                 Relationship='ortholog_one2one')
>>> print orthologs
  Gene(Species='Rattus norvegicus'; BioType='protein_coding'; Description='breast cancer ...

I could also have done that query using a StableId, which I now do using the Ensembl mouse identifier for Brca2.

>>> orthologs = compara.getRelatedGenes(StableId='ENSMUSG00000041147',
...                 Relationship='ortholog_one2one')
>>> print orthologs
  Gene(Species='Rattus norvegicus'; BioType='protein_coding'; Description='breast cancer...

The RelatedGenes object has a number of properties allowing you to get access to data. A Members attribute holds each of the Gene instances displayed above. The length of this attribute tells you how many hits there were, while each member has all of the capabilities described for Gene above, eg. a Cds property. There is also a getSeqLengths method which returns the vector of sequence lengths for the members. This method returns just the lengths of the individual genes.

>>> print orthologs.Members
(Gene(Species='Rattus norvegicus'; BioType='protein_coding'; Descr...
>>> print orthologs.getSeqLengths()
[40742, 84793, 47700]

In addition there’s a getMaxCdsLengths method for returning the lengths of the longest Cds from each member.

>>> print orthologs.getMaxCdsLengths()
[10032, 10257, 9990]

You can also obtain the sequences as a cogent SequenceCollection (unaligned), with the ability to have those sequences annotated as described above. The sequences are named in accordance with their genomic coordinates.

>>> seqs = orthologs.getSeqCollection(feature_types='gene')
>>> print seqs.Names
['Rattus norvegicus:chromosome:12:504...

We can also search for other relationship types, which we do here for a histone.

>>> paralogs = compara.getRelatedGenes(StableId='ENSG00000164032',
...             Relationship='within_species_paralog')
>>> print paralogs
  Gene(Species='Homo sapiens'; BioType='protein_coding'; Description='H2A...

Getting Comparative Alignments

Ensembl stores multiple sequence alignments for selected species. For a given group of species, you can examine what alignments are available by printing the method_species_links attribute of Compara. This will return something like

>>> print compara.method_species_links
Align Methods/Clades
method_link_species_set_id  method_link_id  species_set_id      align_method...
                       788              10           36176             PECAN...
                       780              13           36102               EPO...
                       781              14           36103  EPO_LOW_COVERAGE...

The align_method and align_clade columns can be used as arguments to getSyntenicRegions. This method is responsible for returning SyntenicRegions instances for a given coordinate from a species. As it’s possible that multiple records may be found from the multiple alignment for a given set of coordinates, the result of calling this method is a python generator. The returned regions have a length, defined by the full set of aligned sequences. If the omit_redundant argument is used, then positions with gaps in all sampled species will be removed in the alignment to be returned. The length of the syntenic region, however, is the length of the unfiltered alignment.


It’s important to realise that multiple alignments are from these clades. Hence, sequence regions that you might expect would result in a contiguous alignment in the species subset of interest may be returned as separate SyntenicRegions due to the influence on the alignment of the other species.

>>> syntenic_regions = compara.getSyntenicRegions(region=brca2,
...                      align_method='EPO', align_clade='eutherian')
>>> for syntenic_region in syntenic_regions:
...     print syntenic_region
...     print len(syntenic_region)
...     print repr(syntenic_region.getAlignment(omit_redundant=False))
2 x 1365 dna alignment: Homo sapiens:chromosome:13:32315473-32316418...

We consider a species for which pairwise alignments are available – the bush baby.

>>> compara_pair = Compara(['Human', 'Bushbaby'], Release=Release,
...                        account=account)
>>> print compara_pair
Compara(Species=('Homo sapiens', 'Otolemur garnettii'); Release=81; connected=True)

Printing the method_species_links table provides all the necessary information for specifying selection conditions.

>>> print compara_pair.method_species_links
Align Methods/Clades
method_link_species_set_id  method_link_id  species_set_id      align_method...
                       781              14           36103  EPO_LOW_COVERAGE...
                       682              16           35678         LASTZ_NET...
>>> gene = compara_pair.Bushbaby.getGeneByStableId(
...                             StableId='ENSOGAG00000003166')
>>> print gene
Gene(Species='Otolemur garnettii'; BioType='protein_coding'...
>>> syntenic = compara_pair.getSyntenicRegions(region=gene,
...          align_method='LASTZ_NET', align_clade='H.sap-O.gar')
>>> for region in syntenic:
...     print region
...     break