The work presented in this paper is framed within the context of the BigMed project, a project funded by the Norwegian Research Council. One of the objectives of BigMed is to enhance the phenotype acquisition process in newborns with a monogenetic disorder, one of the four patient groups studied in the project. The use of the Human Phenotype Ontology (HPO) [1] to tag phenotypes and systems like PhenoTips have substantially contributed to the overall phenotype acquisition workflow. PhenoTips [2] is a system for the acquisition of phenotypic information in patients with a genetic disease. PhenoTips also suggests, given a selected set of HPO terms, candidate diagnoses using OMIM (Online Mendelian Inheritance in Man) codes, and related genes for a subsequent genetic test. Although PhenoTips represents a fantastic effort, we believe it could be extended with suitable Semantic Web solutions. In this paper, we present the first steps to adopt ontology alignment techniques to contribute to the diagnostic process.
1 Introduction 2 In order to obtain phenotype-to-disease links between HPO and OMIM terms, as in the reference datasets presented above, we exploited the semantic information in OMIM and the computed mappings. We extracted the associated OMIM disease terms for a given OMIM phenotype term using the ontology properties manifestation of and inheritance type of. The second row in Table 1 shows the number of phenotype-to-disease links derived from the phenotype-to-phenotype mappings.
Results and Discussion. Table 2 shows the coverage of the links derived from the automatically computed alignments.4 The results in terms of Recall are not impressive and only when considering all alignment sets we get near to 50% Recall. Giving a closer look at the results, we noticed that a few OMIM ids present in the annotation files were missing in the OMIM ontology.5 However, the missing OMIM ids are not significant enough to explain the low Recall values. We also observed that a large number (i.e., >1,000) of HPO terms occurring in the links extracted from the annotation files were not aligned to OMIM terms by any of the ontology alignment systems. There may be three possible explanations: (i) the alignment systems fail to discover relevant correspondences, (ii) the OMIM ontology does not cover all phenotype terms from the HPO ontology, and/or (iii) the description of related diseases to a phenotype in the OMIM ontology is limited. Future work aims at clarifying these possible limitations.
Although the results regarding Recall are not encouraging, the results with respect to Precision may be considered promising since there are links derived by the automatically computed correspondences that are not present in the annotation files. These new links, however, require manual curation to assess their validity. 4 The coverage has been calculated in terms of Precision and Recall with respect to the (reference) link sets extracted from the hpo-annotations and phenotips-annotations files. 5 OMIM v.2016AB: https://bioportal.bioontology.org/ontologies/OMIM