OPA2Vec

OPA2Vec generates vector representations of biological entities by combining formal ontology axioms and ontology annotation metadata to improve similarity-based predictions in biomedical research.

Key Features:

• Integration of Ontology Data: Combines formal ontology axioms and annotation axioms/annotation properties, including class labels, descriptions, synonyms, and other metadata.
• Vector Representation Generation: Transforms collected ontology content into feature vectors using a Word2Vec model pre-trained on text corpora such as abstracts or full-text articles.
• Validation and Performance: Demonstrated improved performance compared to existing techniques in benchmark tasks reported by the authors.
• Broad Applicability: Applicable to generating vector representations for diverse biomedical entities derived from various biomedical ontologies and phenotype ontologies.

Scientific Applications:

• Protein-Protein Interaction Prediction: Produces protein vectors that improve similarity measures used to predict protein-protein interactions across datasets.
• Gene-Disease Association Prediction: Generates gene and disease vectors using phenotype ontologies to support prediction of gene-disease associations.
• Candidate Gene Identification from Model Phenotypes: Supports identification of candidate genes for rare and orphan diseases using evidence from mouse model phenotypes.

Methodology:

Combines formal ontology axioms and annotation axioms (annotation properties such as class labels, descriptions, synonyms) and applies a Word2Vec model pre-trained on text corpora (e.g., abstracts or full-text articles) to generate vector representations.