Biomedical Knowledge Graph for Disease-to-Drug Recommendation in Drug Repositioning

This record contains the final six-month biomedical Knowledge Graph used in the dissertation “Knowledge Graph-Based Recommender Systems for Drug Repositioning”. The graph was built for experiments in which a disease is used as a query and candidate drugs are ranked through link prediction. Regulatory drug information was obtained from DailyMed, the Orange Book and the Purple Book and initially processed through MedJsonify. Disease-gene associations were obtained from Open Targets Platform version 26.03, while drug-gene interactions were obtained from DGIdb version 5.0.7. The graph contains 51,759 triples and 10,380 entities: 2,347 drugs, 793 diseases, 2,278 ingredients and 4,962 genes. It includes five relation types: HAS_INGREDIENT, INDICATION, CONTRAINDICATION, ASSOCIATED_WITH_GENE and INTERACTS_WITH_GENE. The package contains the graph in CSV format, entity aliases, the relation schema, quality and entity-resolution reports, source and build information, checksums, and the results of an audit of drug-name resolution cases. This dataset can be used to study biomedical Knowledge Graph construction and drug-disease link prediction. A relation in the graph or a high score produced by a model does not demonstrate clinical efficacy, safety or a new therapeutic indication. Model outputs require independent biomedical validation.