Fig. 5

Disease gene discovery via semantic similarity and protein–protein interaction network. a Our semantically driven disease gene discovery approach using external omic knowledge. This approach establishes the semantic neighborhood of a patient to identify a relevant known disease gene set, and then recruits prior knowledge of relevant gene–gene relationships to intersect with patient variations. This integration of the disease catalog with omic knowledge results in potential variant discovery and phenotypic extension of known disease genes. As shown in this example, a patient phenotype query determines training genes: those variant in the patient and known to contain variants causing cataloged diseases most similar to the query. The biological subnetwork implicated by these training genes is then realized in “omic space.” For this example, proteomic space, as defined by the PINA2 protein–protein interaction network, is used. This process identifies candidate genes that are variant in the patient and connected to training genes in the protein interaction network. In this figure, an additional constraint has been applied, in which genes must directly interact with at least two training genes to be considered candidates. b To validate this procedure we performed a global analysis across the entirety of Online Mendelian Inheritance in Man (OMIM). In protein interaction network space, the variant genes of nearest semantic neighbor diseases are typically closer to each other than to those of all diseases