Onto2Vec

UPDATE

An improved version of our method is available at : https://github.com/bio-ontology-research-group/opa2vec/

Introduction

Onto2Vec is a program that can be used to produce feature vectors for biological entities based on their annotations to biomedical ontologies. Onto2Vec uses the integrity of the axioms describing the structure of an ontology as well as the entity-concept annotation axioms. This is a detailed documentation on how to use Onto2Vec to produce ontology-based feature vectors. The Onto2Vec implementation shared here uses GO ontology as a case study and can be directly used to produce feature vectors for GO functions and proteins. However, Onto2Vec can also be used to produce feature vectors for entities using any other biomedical ontology. We provide here instructions on how to use Onto2Vec with the ontology of your choice.

Prerequisites

Onto2Vec implementation uses Groovy with Grape for dependency management (http://docs.groovy-lang.org/latest/html/documentation/grape.html), Python and Perl. No other programs are required to run it.

OWL Processing

The first step of Onto2Vec is to process the ontology using OWL API and infer new axioms using a reasoner.

• Create a new directory and name it Onto2Vec.
• In this directory paste the file OntoProcess.groovy.
• If you want to use the GO ontology:
  • Open the terminal and run

    groovy OntoProcess.groovy
    

  The script should create two files in your directory: classes.lst which contains the set of classes in the ontology, and axioms.lst which contains all axioms from the ontology including the inferred ones.
• If you want to use a different ontology:
  • Open the file OntoProcess.groovy in editor.
  • Edit the following line:

    IRI iri= IRI.create ("http://purl.obolibrary.org/obo/go.owl");
    

  by replacing the url with the link to the owl version of the ontology you would like to use.
  • Please note that the file uses the reasoner to automatically infer axioms of type SubClass, Equivalent and Disjoint. You may need to change that depending on the axioms available in your ontology and what type of axioms you would like to infer.

Adding Annotation axioms

The second step of Onto2Vec is to add annotation axioms to our corpus of axioms. This can be easily done using some simple text mining code without using the OWL API sice there is no needed owl processing.

• If you are using GO ontology you can add the annotation axioms by following these steps:
  • Download the GOA annotation file from ftp://ftp.ebi.ac.uk/pub/databases/GO/goa/UNIPROT/ and save it to the Onto2Vec directory. For copyright reasons, we do not include the annotation file in this package.
  • Save the file getAnnotations.pl in the Onto2Vec directory.
  • Open the file getAnnotations.pl in editor.
  • Edit the following line:

    my $database="Enter Path Here";
    

    by replacing Enter Path Here with the path to the annotation file you downloaded.
  • Open the terminal and run

    perl getAnnotations.pl 
    

    This script should create a file called annotationAxiom.lst with all experimental annotation axioms using hasFunction as the relation between proteins and their functions .
  • To make sure the annotation axioms are complete, we need to propagate the annotations up in the GO DAG.To do that save the file AddAncestors.pl to the Onto2Vec directory.
  • Open the terminal and run

     perl AddAncestors.pl
    

  A file annotationAxiomFinal.lst should now be created and should include all the needed annotation axioms.
  • At this point, the GO ontology axioms and the annotation axioms needs to be merged into one corpus. To do that, open the terminal in the Onto2Vec directory and run:

    cat axioms.lst annotationAxiomFinal.lst > AllAxioms.lst
    

• If you are using a different biomedical ontology:
  • If you do not need to add annotation axioms (Onto2Vec can work without annotation axioms)
    • Open the terminal in the Onto2Vec directory and run

    cat axioms.lst  > AllAxioms.lst
    

  • If you would like to include the annotation axioms, do the following:
    • Upload and save the annotation file corresponding to the ontology you are using in the Onto2Vec directory.
    • Open the file getAnnotations.pl in editor.
    • Edit the following line:

    my $database="Enter Path Here";
    

    by replacing Enter Path Here with the path to the annotation file you downloaded.
    • Edit the line :

    if ($line !~"	IEA" and $line !~"	ND" )
    

    by replacing IEA and ND with the evidence codes you want to filter (if there are any). Or you can comment the if statement and get all the annotaions in the file.
    • Open the terminal and run

    perl getAnnotations.pl 
    

    The obtained file annotationAxiom.lst should have the annotation axioms you added.
    • To merge the ontology axioms with the annotation axioms in one corpus, open the terminal and run

     cat axioms.lst annotationAxiom.lst > AllAxioms.lst
    

Representation Learning

Since the axioms corpus is now ready, the only step left is to run Word2Vec to obtain the representation vectors in two steps:

• The first step is to obtain the entities we would like to get the feature vectors for. To get the vectors for all classes, copy and paste getClasses.pl in the Onto2Vec directory and then open the terminal and run

  perl getClasses.pl
  

  This script will create a file, AllClasses.lst, with the list of all classes in it. You can easily edit the perl file getClasses.pl to only keep the classes you are interested in such as only the annotated entities or so.
• The second and last step is to run Word2Vec on the corpus. To do that, copy and paste runWord2Vec.py in the Onto2Vec directory and open the terminal and run

  python runWord2Vec.py
  

  This script will create a file with the vector representations of all classes in AllClasses.lst. This file, VecResults.lst, is the final and main output of Onto2Vec.

Final notes

For any comments or help needed with how to run Onto2Vec, please send an email to: [email protected]