Fig. 7.

Dug versus lexical search performance using UMLS synonyms of TOPMed phenotype concepts to search for underlying variables. Left: Recall, R@10, R@50. Right: Precision, P@10, P@50. Diamonds represent outliers for each search method. Dug significantly outperforms lexical search (implemented in Elasticsearch) in terms of total recall, R@10 and R@50. Precision @10 and P@50 were not significantly different between Dug and Elasticsearch. Though Dug’s total precision was significantly lower, this was expected given Dug’s objective to uncover exploratory connections