Abstract

Strategies for predicting gene-disease connections based on ontology cover a wide range of techniques, from conventional semantic similarity algorithms to more contemporary developments such as knowledge graph embeddings. While semantic similarity often delves within the hierarchical relations embedded in the ontology, knowledge graph embeddings take a broader view, considering a multitude of connections. Nevertheless, these embeddings are usually constructed across a single network, and additional ontologies would be necessary for more complicated tasks like gene-disease correlation prediction. The proposed research is delved into the impact of utilizing richer semantic representations derived from multiple ontologies. These representations aim to capture a comprehensive understanding of both genes and diseases, taking into account diverse types of relationships within the ontologies. The utility of using random walk-based knowledge graph embeddings, emphasizes the need for close integration of different ontologies to improve prediction performance. The multi-ontology approach contributes to a more nuanced and holistic interpretation of gene-disease associations, paving the way for improved biomedical insights.