Figure 5
Knowledge graph construction. (A) A heterogeneous knowledge graph is constructed from literature evidence, compound libraries, experimental data, databases of collated and curated datasets, and clinical trial evidence (B) A canonical set of connections in the graph. Note that these edges can be directional: 1 GeneProtein entities are connected to other GeneProtein entities through protein-protein interactions (PPI). 2 GeneProteins are connected to Diseases through Therapeutic Relationships (benchmark) and Biological Association relations. 3 GeneProteins are connected to Pathways via Biological Association relations, and are connected to GO processes via a Biological Association link and a Therapeutic Link relation. 4 Compounds are connected to Diseases via Therapeutic Links. 5 GeneProteins are connected to compounds via Experimental Evidence edges. 6 Diseases are connected to Pathways and GO processes via a Mechanistic Connection relation. 7 GO processes are connected to GO processes via Biological Association relations. (C) The percentage compositions of the five biomedical entities is shown here, with Compounds dominating the graph. (D) The number of edges between each pair of entities. GP=GeneProtein, GO=GO process, D=Disease, C=Compound, P=Pathway. Shown in red is the benchmark relation, GeneProtein-Disease Therapeutic Relationship.
