The Unified Medical Language System (UMLS) integrates knowledge from several biomedical terminologies. One of the important sources of knowledge is the set of semantic relationships asserted between the concepts derived from different source terminologies. Such integration of multiple sources of terminological knowledge can potentially enable discovery of implicit meaningful relationships between the concepts (unrelated at a given time).
In this research we explore whether the existing knowledge in the UMLS in the form of semantics (types or categories) and structure (network topology) can be used to discover potential relationships. In this paper, we propose a problem-independent approach to discover potential terminological relationships by semantic abstraction of transitive relationship paths to perform classification and analyzing network theoretic measures such as topological overlap and average number of transitive paths. Using different versions of the UMLS, we evaluate the proposed approach over newly added relationships.
Strong discriminative characteristics were observed with semantic abstraction based classifier (area under curve of 0.96) and average number of transitive paths (t[198]=−4.19, p<0.001) to identify potential relationships.
The UMLS has sufficient knowledge to enable discovery of potential terminological relationships.