Abstract:
This work shows an overall method of using more than one biomedical knowledge graph to improve learning with graphs in the field of biomedical study. We combined various knowledge graphs with entities like proteins, genes, diseases, metabolites, phenotypes and chemicals. At first, we made the data uniform by standardizing it to increase its compatibility and consistency. After that, we set up a graph database using Neo4j. This made it easier to manage and get data from the system. We then assessed different Knowledge Graph Embedding (KGE) models and Graph Neural Network (GNN) structures to measure their performance in downstream tasks - mainly link prediction. By doing multiple tests, we showed how our method can effectively use the combined knowledge stored in biomedical knowledge graphs for better predictive modelling and finding new information in biomedical research fields.
