Following the steps on this page, you will set up an example InterMine. You will:
- Load some real data sets for Malaria (P. falciparum)
- Learn about how data integration works
- Deploy a webapp to query the data
Note See Quick Start if you are impatient and just want to run an InterMine.
We use git to manage and distribute source code and gradle as our build system. InterMine makes use of a variety of freely available software packages. For this tutorial, you will need the following software packages installed locally and running:
|Git||1.7||It is our source control software. We use it to check out, update, manage, and distribute our source code. Note: InterMine is available via Maven Central as executable JARs. We do not recommend downloading the InterMine source code.|
|Java SDK||8||We use Gradle as our build system and the usage of Gradle requires a Java Software Development Kit (Java SDK) installation. We recommend you use OpenJDK as it’s probably safer moving forward.|
|PostgreSQL||9.3.x||It is a powerful, open source object-relational database system that uses and extends the SQL language combined with many features that safely store and scale the most complicated data workloads. We use it for our database.|
|Tomcat||8.5.x||It is an open source implementation of the Java Servlet, JavaServer Pages, Java Expression Language and Java WebSocket technologies. We use it for deploying the web application.|
|Solr||7.2.1||Solr makes it easy for programmers to develop sophisticated, high-performance search applications with advanced features. We use it for its keyword search in our search engines.|
|Perl||5.8.8||Many of the build processes are carried out by Perl programs. You will not need it for this tutorial, however, you will need it installed on your system to build or maintain an InterMine installation.|
Note: InterMine only supports installations onto Linux and Mac OS X systems. Windows systems are not supported. We run a mixture of Debian and Fedora servers in our data centre in Cambridge. See Software for configuration details.
Download the mine code from GitHub.
Gradle has helper processes enabled by default. We're going to disable those by setting
If at any point you need help or have a quick (or not so quick) question, please get in touch! We have a discord server, twitter and a developer mailing list.
BioTestMine is a dummy test mine we use to test out new features which contains real (old) data for Malaria (P. falciparum).
To get started, change into the directory you checked out the BiotestMine source code to and look at the sub-directories:
|/dbmodel||contains information about the data model and related configuration files|
|/webapp||basic configuration for the webapp|
|/data||contains a tar file with data to load|
|build.gradle||The –stacktrace option will display complete error messages if there is a problem.|
|gradle.properties||Sets system variables. Determines which version of InterMine you use.|
|settings.gradle||Sets gradle projects. Do not edit.|
|project.xml||Configures which data parsers are run during your build.|
There is also a gradle directory (
/gradle) and executables (
project.xmlallows you to configure which data to load into your Mine. The file has two sections: sources and post-processing.
<source> elements list and configure the data sources to be loaded. A source can have a name and a type.
Corresponds to the name of the bio-source artifact (jar) which includes parsers to retrieve data and information on how it will be integrated.
Can be anything and can be the same as
type. Using a more specific name allows you to define specific integration keys.
<source> elements can have several properties depending on source type.
src.data.includes are all used to define locations of files that the source should load. Other properties are used as parameters to specific parsers.
Specific operations can be performed on the Mine once data is loaded, these are listed here as
<post-process> elements. We will look at these in more detail later.
The biotestmine git repository includes a tar file with data to load into BiotestMine. These are real, complete data sets for P. falciparum (but very old!).
We will load genome annotation from PlasmoDB, protein data from UniProt and GO annotation also from PlasmoDB.
See Data files to integrate for details on the data.
Copy this to a local directory (your home directory is fine for this workshop) and extract the archive:
A dummy project XML file is available in the
/data/ directory. Copy it into your
biotestmine directory, then edit
project.xml to point each source at the extracted data, just replace
/home/username (or on a mac,
/Users/username). Do use the absolute path.
For example, the
Note All file locations must be absolute not relative paths.
project.xml file is now ready to use.
Configuration of local databases and tomcat deployment is kept in a
MINE_NAME.properties file in a
.intermine directory under your home directory. We need to set up a
If you don't already have a
.intermine directory in your home directory, create one now:
There is a partially completed properties file for BioTestMine already. Copy it into your
Update this properties file with your postgres server location, username and password information for the two databases you just created. The rest of the information is needed for the webapp and will be updated later.
For the moment you need to change
PSQL_PWD in the
If you don't have a password for your postgres account you can leave
Finally, we need to create
items-biotestmine postgres databases as specified in the
New postgres databases default to
UTF-8 as the character encoding. This will work with InterMine but performance is better with
Now we're ready to set up a database schema and load some data into our BioTestMine. First, some information on how data models are defined in InterMine.
InterMine uses an object-oriented data model. Classes in the model and relationships between them are defined in an XML file. Depending on which data types you include, you will need different classes and fields in the model, so the model is generated from a core model XML file and any number of
additions files. These additions files can define extra classes and fields to be added to the model.
Elements of the model are represented by Java classes and references
These Java classes map automatically to tables in the database
The object model is defined as an XML file, that defines
The Java classes and database schema are automatically generated
from an XML file.
The model is generated from a core model XML file and any number of additions files defined in the dbmodel/build.gradle file.
The core InterMine data model is defined in core.xml file.
Note the fields defined for
Protein is a subclass of
BioEntity, defined by
Protein class will therefore also inherit all fields of
The first file merged into the core model is the
so_additions.xml file. This XML file is generated in the
dbmodel/build directory from terms listed in the
so_terms file, as configured in the dbmodel/build.gradle file.
The build system creates classes corresponding to the Sequence Ontology terms.
The model is then combined with any extra classes and fields defined in the sources to integrate, i.e. those listed as
<source> elements in
project.xml. Look at the additions file for the UniProt source, for example. This defines extra fields for the
Protein class which will be added to those from the core model.
Now run the gradle task to merge all the model components, generate Java classes and create the database schema:
The clean task is necessary when you have run the task before, it removes the
build directory and any previously generated models.
This task has done several things:
Merged the core model with other model additions and created a new
XML file:~/git/biotestmine $ less dbmodel/build/resources/main/genomic_model.xml
Looking through the
Proteinclass, you can see it combines fields from the core model and the UniProt additions file.
so_additions.xmlfile has also been created using
the sequence ontology terms in
so_term:~/git/biotestmine $ less dbmodel/build/so_additions.xml
Each term from
so_termwas added to the model, according to the sequence ontology.
Generated and compiled a Java class for each of the
<class>elements in the file. For example
Protein.java:~/git/biotestmine $ less dbmodel/build/gen/org/intermine/model/bio/Protein.java
Each of the fields has appropriate getters and setters generated for it. Note that these are
interfacesand are turned into actual classes dynamically at runtime - this is how the model copes with multiple inheritance.
Automatically created database tables in the postgres database specified in
db.production- in our case
Log into this database and list the tables and the columns in the
protein table:$ psql biotestminebiotestmine=# \dbiotestmine=# \d protein
The different elements of the model XML file are handled as follows:
There is one column for each attribute of
Protein - e.g.
References to other classes are foreign keys to another table - e.g.
Protein has a reference called
organism to the
Organism class so in the database the
protein table has a column
organismid which would contain an id that appears in the
Indirection tables are created for many-to-many collections - e.g.
Proteinhas a collection of
Geneobjects so an indirection table called
genesproteins is created.
This has also created necessary indexes on the tables:
buildDB will destroy any existing data loaded in the biotestmine database and re-create all the tables.
The model XML file is stored in the database once created. This and some other configuration files are held in the
intermine_metadatatable which has
Now, we have the correct data model and the correct empty tables in the database. We can now run several data parsers to load our data into our database.
For this tutorial, we will run several data integration and post-processing steps manually. This is a good way to learn how the system works and to test individual stages. For running actual builds there is a
project_build script that will run all steps specified in
project.xml automatically. We will cover this later.
Loading of data is done by running the
integrate gradle task.
|./gradlew||Use the provided gradle wrapper so that we can be sure everyone is using the same version.|
|integrate||Gradle task to run the specified data source|
|-Psource=||Data source to run. Source name should match the value in your project XML file|
|-stacktrace||The –stacktrace option will display complete error messages if there is a problem.|
This will take a couple of minutes to complete, the command runs the following steps:
Checks that a source with name
Reads the UniProt XML files at the location specified by
Runs the parser included in the UniProt JAR. The JARs for every core
InterMine data source are published in Maven Central. The build
looks for jar with the name matching "bio-source-<source-type>-<version>.jar", e.g.
bio-source-uniprot-2.0.0.jar. Maven will automatically download the correct JARs for you.
The UniProt data parser reads the original XML and creates
which are metadata representations of the objects that will be loaded into the biotestmine database. These items are stored in an intermediate
itemsdatabase (more about
Reads from the
itemsdatabase, converts items to objects and loads them into the biotestmine database.
This should be completed after a couple of minutes. Now that the data has loaded, log into the database and view the contents of the protein table:
And see the first few rows of the data:
InterMine works with objects. Objects are loaded into the production system and queries return lists of objects. These objects are persisted to a relational database. Internal InterMine code (the ObjectStore) handles the storage and retrieval of objects from the database automatically. By using an object model, InterMine queries benefit from inheritance, for example, the
Exonclasses are both subclasses of
SequenceFeature. When querying for SequenceFeatures (representing any genome feature) both Genes and Exons will be returned automatically.
We can see how inheritance is represented in the database:
One table is created for each class in the data model.
Where one class inherits from another, entries are written to both
tables. For example:
The same rows appear in the
All classes in the object model inherit from
InterMineObject. Querying the
intermineobject table in the database is a useful way to find the total number of objects in a Mine:
All tables include an
id column for unique ids and a
class column with the actual class of that object. Querying the
class column of
intermineobject, you can find the counts of different objects in a Mine:
A technical detail: for speed when retrieving objects and to deal with inheritance correctly (e.g. to ensure a
Gene object with all of its fields is returned even if the query was on the
SequenceFeature class), a serialised copy of each object is stored in the
intermineobject table. When queries are run by the ObjectStore, they actually return the ids of objects - these objects may already be in a cache, if not, the are retrieved from the
We will load genome annotation data for P. falciparum from PlasmoDB
- genes, mRNAs, exons and their chromosome locations - in GFF3 format
- chromosome sequences - in FASTA format
Note that genes from the GFF3 file will have the same
primaryIdentifier as those already loaded from UniProt. These will merge in the database such that there is only one copy of each gene with information from both data sources. We will load the genome data and then look at how data integration in InterMine works.
First, look at the information currently loaded for gene
PFL1385c from UniProt:
GFF3 is a standard format used to represent genome features and their locations. Each line represents one feature and has nine tab-delimited columns as shown below:
col 1: "seqid"
an identifier for a 'landmark' on which the current feature is locatated, in this case 'MAL1', a ''P. falciparum'' chromosome.
col 2: "source"
the database or algorithm that provided the feature
col 3: "type"
a valid Sequence Ontology term defining the feature type - here
col 4 & 5: "start" and "end"
coordinates of the feature on the landmark in col 1
col 6: "score"
an optional score, used if the feature has been generated by an algorithm
col 7: "strand"
'+' or '-' to indicate the strand the feature is on
col 8: "phase"
CDS features to show where the feature begins with reference to the reading frame
col 9: "attributes"
custom attributes to describe the feature. These are name/value pairs separated by ';'. Some attributes have predefined meanings, relevant here:
ID- identifier of feature, unique in scope of the GFF3 file
Name- a display name for the feature
IDof another feature in the file that is a parent of this one. In our example the
A dot means there is no value provided for the column.
The files we are loading are from PlasmoDB and contain
mRNA features. There is one file per chromosome. Look at an example:
InterMine includes a parser to load valid GFF3 files. The creation of features, sequence features, locations and standard attributes is taken care of automatically.
gff3 properties can be configured in the
project.xml. The properties set for
gff3.seqClsName = Chromosome
the ids in the first column represent
Chromosome objects, e.g. MAL1
gff3.taxonId = 36329
taxon id of malaria
gff3.dataSourceName = PlasmoDB
the data source for features and their identifiers. This is used for the DataSet (evidence) and synonyms.
gff3.seqDataSourceName = PlasmoDB
the source of the seqids (chromosomes) is sometimes different to the features described
gff3.dataSetTitle = PlasmoDB P. falciparum genome
a DataSet object is created as evidence for the features, it is linked to a DataSource (PlasmoDB)
You can also configure GFF properties in the gff.config file. See GFF3 for details.
To deal with any specific attributes or perform custom operations on each feature, you can write a handler in Java which will get called when reading each line of GFF. For malaria gff, we need a handler to switch which fields from the file are set as
secondaryIdentifier in the features created. This is to match the identifiers from UniProt, it is quite a common issue when integrating from multiple data sources.
From the example above, by default:
ID=gene.46311;description=hypothetical%20protein;Name=PFA0210c would make
PFA0210c. We need
PFA0210c to be the
Look at the
malaria-gff.properties file - there are two properties of interest:
The property file has specified a Java class to process the GFF file, MalariaGFF3RecordHandler. This code changes which fields the
Name attributes from the GFF file have been assigned to.
Now execute the
malaria-gff source by running this command:
This will take a few minutes to run. Note that this time, we don't run
buildDB since we are loading this data into the same database as UniProt. As before, you can run a query to see how many objects of each class are loaded:
FASTA is a minimal format for representing sequence data. Files comprise a header with some identifier information preceded by '>' and a sequence. At present, the InterMine FASTA parser loads just the first entry in header after
> and assigns it to be an attribute of the feature created. Here we will load one FASTA file for each malaria chromosome. Look at an example of the files we will load:
The type of feature created is defined by a property in
project.xml, the attribute set defaults to
primaryIdentifier, but can be changed with the
fasta.classAttribute property. The following properties are defined in
fasta.className = org.intermine.model.bio.Chromosome
the type of feature that each sequence is for
fasta.dataSourceName = PlasmoDB
the source of identifiers to be created
fasta.dataSetTitle = PlasmoDB chromosome sequence
a DataSet object is created as evidence
fasta.taxonId = 36329
the organism id for malaria
fasta.includes = MAL*.fasta
files to process
This will create features of the class
primaryIdentifier set and the
Chromosome.sequence reference set to a
Sequence object. Also created are a
DataSource as evidence.
Now run the
malaria-chromosome-fasta source by running this command:
This has integrated the chromosome objects with those already in the database. In the next step, we will look at how this data integration works.
malaria-gff have both loaded information about the same genes. Before loading genome data, we ran a query to look at the information UniProt provided about the gene "PFL1385c":
Which showed that UniProt provided
symbol attributes and set the
organism reference. The
id was set automatically by the ObjectStore and will be different each time you build your Mine.
Running the same query after
malaria-gff is added shows that more fields have been filled in for same gene and that it has kept the same id:
This means that when the second source was loaded, the integration code was able to identify that an equivalent gene already existed and merged the values for each source. The equivalence was based on
primaryIdentifier as this was the field that the two sources had in common.
malaria-gff does not include a value for
symbol but it did not write over the
symbol provided by UniProt. Actual values always take precedence over null values (unless configured otherwise).
Now look at the organism table:
Three sources have been loaded so far that all included the organism with
taxonId 36329, and more importantly, they included objects that reference the organism. There is still only one row in the organism table implying that the data from the three sources has merged, in this case
taxonId was the field used to define equivalence.
Data integration works by defining keys for each class of object to describe fields that can be used to define equivalence for objects of that class. For the examples above:
primaryIdentifierwas used as a key for
taxonIdwas used as a key for
Gene object loaded by
malaria-gff, a query was performed in the
biotestmine database to find any existing
Gene objects with the same
primaryIdentifier. If any were found, fields from both objects were merged and the resulting object stored.
Many performance optimisation steps are applied to this process. We don't actually run a query for each object loaded, requests are batched and queries can be avoided completely. If the system can work out, no integration will be needed.
We may also load data from some other source that provides information about genes but doesn't use the identifier scheme we have chosen for
primaryIdentifier (in our example
PFL1385c). Instead it only knows about the
ABRA), in that case, we would want that source to use the
symbol to define equivalence for
keydefines a field or fields of a class that can be used to search for equivalent objects.
- Multiple primary keys can be defined for a class. Sources can use different keys for a class if they provide different identifiers
- One source can use multiple primary keys for a class if the objects of that class don't consistently have the same identifier type.
null- if a source has no value for a field that is defined as a primary key, then the key is not used and the data is loaded without being integrated.
The keys used by each source are set in the source's
The key on
Gene.primaryIdentifier is defined in both sources, that means that the same final result would have been achieved regardless of the order in which the two sources were loaded.
_keys.properties files define keys in the format:
name_of_key can be any string but you must use different names if defining more than one key for the same class, for example in
uniprot_keys.properties, there are two different keys defined for
It is better to use common names for identical keys between sources as this will help avoid duplicating database indexes. Each key should list one or more fields that can be a combination of
attributes of the class specified or
references to other classes, in this case, there should usually be a key defined for the referenced class as well.
tracker table is created in the target database by the data integration system. This tracks which sources have loaded data for each field of each object. The data is used along with priorities configuration when merging objects but is also useful to view where objects have come from.
- Look at the columns in the tracker table;
objectidreferences an object from some other table
- Query tracker information for the objects in the examples above:
Organisms and publications in InterMine are loaded by their taxon id and PubMed id respectively. The
update-publications sources can be run at the end of the build to examine the ids loaded, fetch details via the NCBI Entrez web service and add those details to the Mine.
You will have noticed that in previous sources and in
project.xml, we have referred to organisms by their NCBI Taxonomy id. These are numerical ids assigned to each species. We use these for convenience in integrating data. The taxon id is a good unique identifier for organisms, whereas names can come in many different formats. For example, in fly data sources we see:
D. melanogaster, Dmel, DM, etc.
Looking at the
organism table in the database, you will see that the only column filled in is
From the root
biotestmine directory run the
This should only take a few seconds. This source does the following:
runs a query in the production database for all of the taxon ids
creates an NCBI Entrez web service request to fetch details of those
converts the data returned from Entrez into a temporary Items XML
loads the Items XML file into the production database
Now run the same query in the production database, you should see details for ''P. falciparum'' added:
As this source depends on organism data previously loaded, it should be one of the last sources run and should appear at the end of
Publications are even more likely to be cited in different formats and are prone to errors in their description. We will often load data referring to the same publication from multiple sources and need to ensure those publications are integrated correctly. Hence, we load only the PubMed id and fetch the details from the NCBI Entrez web service as above.
Several InterMine sources load publications:
Now run the
update-publications source to fill in the details:
As there are often large numbers of publications, they are retrieved in batches from the web service.
Now details will have been added to the
As this source depends on publication data previously loaded, it should be one of the last sources run and should appear at the end of
Post-processing steps are run after all data is loaded, they are specified as
<post-process> elements in
Some of these can only be run after data from multiple sources are loaded. For example, for the Malaria genome information, we load features and their locations on chromosomes from
malaria-gff but the sequences of chromosomes from
malaria-chromosome-fasta. These are loaded independently and the
Chromosome objects from each are integrated. Neither of these on their own could set the sequence of each
Exon. However, now that they are both loaded, the
transfer-sequences post-process can calculate and set the sequences for all features located on a
Chromosome for which the sequence is known.
Some post-process steps are used to homogenize data from different sources or fill in shortcuts in the data model to improve usability - e.g.
Finally, there are post-process operations that create summary information to be used by the web application:
<post-process> targets are included in the BioTestMine
Run queries listed here before and after running the post-processing steps to see examples of what each step does.
This fills in some shortcut references in the data model to make querying easier. For example,
Gene has a collection of
Transcript has a collection of
create-references will follow these collections and create a
gene reference in
Exon and the corresponding
exons collection in
geneid column will be filled in, representing the reference to gene.
create-references postprocess by running this command:
The sequence for chromosomes is loaded by
malaria-chromosome-fasta, but no sequence is set for the features located on them. This step reads the locations of features, calculates and stores their sequence and sets the
sequenceid column. The
sequenceid column for this exon is empty:
transfer-sequences postprocess by running this command:
sequenceid column is filled in.
Each source can also provide code to execute post-process steps if required. This command loops through all of the sources and checks whether there are any post-processing steps configured. There aren't any for the sources we are using for BioTestMine but you should always include the
These generate summary data and search indexes used by the web application, see Keyword Search for details.
summarise-objectstore postprocess by running this command:
You must have Solr installed and running for the indexes to be populated correctly.
Download Solr binary package and extract it to any place you like. Inside
/solr-7.2.1 directory start the server with this command:
Initialising Search Indexes
To create a Intermine collection for search process, run this command inside the solr directory.
To create a Intermine collection for autocomplete process, run this command inside the solr directory.
These are empty search indexes that will be populated by the
See Solr for details.
create-autocomplete-index postprocesses by running these commands:
So far, we have created databases, integrated data and run post-processing with individual gradle tasks. Alternatively InterMine has a Perl program called
project_build that reads the
project.xml definition and runs all of the steps in sequence. The script has the option of creating snapshots during the build at specified checkpoints.
To build BioTestMine using the
project_build script, first download the script:
project_build script from your
This will take ~15-30mins to complete.
Note If you encounter an "OutOfMemoryError", you should set your $GRADLE_OPTS variable. See Troubleshooting tips.
You can deploy a web application against your newly built database.
~/.intermine directory, update the webapp properties in your biotestmine.properties file. Update the following properties:
- tomcat username and password
- superuser username and password
The userprofile database stores all user-related information such as username and password, tags, queries, lists and templates. To build the userprofile database:
Update your biotestmine.properties file with correct information for the
Create the empty database:$ createdb userprofile-biotestmine
Build the database:# creates the empty tables~/git/biotestmine $ ./gradlew buildUserDB
You only need to build the userprofile database once.
Warning The buildDB and buildUserDB commands rebuild the database and thus will delete any data.
Before deploying the biotestmine webservices, you need to configure tomcat. See Tomcat for configuration details.
Run the following command to deploy your webapp:
If you make changes, redeploy your webapp with this command:
BlueGenes is the new user interface. It runs as its own service and utilises the InterMine web service API.
Run the following command to start BlueGenes:
This approach is only recommended for trying out the app. See BlueGenes for deploying to a production environment.
You should be able to access the new user interface BlueGenes from http://localhost:5000
If you want to use the legacy user interface visit http://localhost:8080/biotestmine. The path to your webapp is the
webapp.path value set in biotestmine.properties.
Now that you have a database and a working webapp, you'll want to know how to add your own logo, pick a colour scheme, modify how data is displayed etc. In the Web Application section you'll find a detailed guide on how to customise all parts of the InterMine web application.
Anytime you run
./gradlew and something bad happens, add the
This will give you a more detailed output and hopefully a more helpful error message.
If the error occurs while you are integrating data, the error message will be in the
intermine.log file in the directory you are in.
If the error occurs while you are browsing your webapp, the error message will be located in the Tomcat logs:
Please contact us if you run into problems. We have a discord server, twitter and a developer mailing list.