Ontology embeddings of HPO, ORDO, and HOOM

This describes the ontology embeddings of HPO, ORDO, and HOOM for training Onto-CGAN (Paper: Generating Unseen Diseases Patient Data Using Ontology-enhanced Generative Adversarial Networks.)We first combined ORDO, HPO, and HOOM ontologies to create unified resource-capturing relationships between diseases, phenotypes, and other biomedical concepts. The combined ontology is processed by OWL2Vec* to transform into graph-based representations, where nodes represent concepts (e.g., diseases, phenotypes) and edges represent their relationships. We used a random walk method with a depth of 3 to explore the graph structures to capture semantic relationships. Textual annotations (e.g., definitions) are tokenized and processed through a word2vec model. The word2vec model was trained by 10 iterations with a window size of 5 words and a minimum word frequency of 1. The resulting embeddings are 100-dimensional vectors that integrate hierarchical relationships, logical axioms, and textual information from the combined ontology. The embedding file is <i>ontology.embeddings </i>and the dictionary files are <i>HPO.dict</i> and <i>ORDO.dict</i>