Fig. 1.

Illustration of the approaches that we used to calculate phenotypic similarity. Resnik's similarity uses the taxonomy of the ontology. OPA2Vec generates vector representations by using the axioms of the ontologies propagated over the subsumption hierarchy along with the natural language information available in the ontology. DL2Vec and OWL2Vec generate a graph from the ontologies axioms then perform random walks to generate vector representations for genes and diseases, with some differences including that the graphs are directed in OWL2Vec and undirected in DL2Vec.