mOWL

mOWL implements machine learning methods that integrate biomedical ontologies formalized in the Web Ontology Language (OWL) into vector-space representations for downstream bioinformatics analyses.

Key Features:

• Ontology Embedding: Implements methods to embed OWL ontologies and formal knowledge bases into vector spaces while preserving semantic relations.
• Similarity Computation: Computes semantic similarity using embedded ontological representations to compare biological entities.
• Deductive Inference: Supports deductive reasoning over OWL ontologies to derive logical consequences from formal axioms.
• Zero-Shot Learning: Enables zero-shot prediction for unseen classes or entities by leveraging ontology embeddings and semantic relations.

Scientific Applications:

• Knowledge-Based Prediction of Protein-Protein Interactions: Uses the Gene Ontology and ontology-derived embeddings to support prediction of protein–protein interactions based on functional annotations.
• Gene-Disease Associations: Employs phenotype ontologies and semantic representations to identify associations between genes and diseases.

Methodology:

Computational methods explicitly include embedding OWL ontologies into vector spaces while preserving semantic relationships, similarity computation on embeddings, deductive inference over ontologies, and zero-shot prediction using semantic representations.