Leading ontology alignment (OA) systems add background knowledge sources to find mappings that string-based matchers are unable to find. Mediating ontologies (MOs) are typically pre-selected before the OA process begins. BioPortal Mediating Finder was developed to use BioPortal as a repository to dynamically select and access MOs for OA. BioPortal [1] provides access to over 370 biomedical ontologies and their mappings. The software development and results on the OAEI anatomy and large biomedical tracks with LogMap and several OA systems are reported. In [2] the fastselection approach to find MOs was presented. Experiments produced the top five: Mouse Anatomy, SYN, Uberon, CL and EHDAA2 for the anatomy track. The following algorithm is used with each of these mediating ontologies.
LogMap+ Bioportal Mediating Finder (2 MOs AND Cf >0.7) OR (1 MO AND Cf > 0.8)
0.9125 0.8793 Experimental results on the large biomedical track with the two versions of each ontology are in Table 2. The top five mediating ontologies were determined using the fastselection approach. The top four MOs for the anatomy track were in the top four MOs for the alignment task with the FMA. Only SYN was in the top five for SNOMEDNCI tasks. Table 2 shows results first for the small versions and then for the complete ontologies. LogMap is compared to that of Logmap + BMF.
Comparing LogMap to LogMap + BioPortal Mediating Finder (BMF) SMALL
0.8793 0.9030 + BMF 0.7443 0.8570 0.7967
0.9643 0.6680 0.7892 0.8753 0.5967 0.7096 + BMF +BMF Three (in bold) have improved f-measures. The large number of concepts in FMA and NCI result in many new mappings, significantly reducing precision (in italics) although recall increases. More sophisticated filtering is needed as in the anatomy track. BMF with other recent OAEI systems, two top (AML and GOMMA) and two poorer (Hertuda and AOT) is in Table 3. F-measures (bold) increases though slightly for Hertuda.
AML + BMF Hertuda + BMF Prec 0.9548 0.9319 0.6892 0.7024