Introduction
Context: Created in 1986 by the U.S. National Library of Medicine, the Unified
Medical Language System (UMLS®)[
1
] is a database that regroups numerous
vocabularies and biomedical health norms so as the facilitate interoperability between
informatics systems. The UMLS is used to ameliorate or develop applications such as
Electronic Health Record’s classification tools, dictionaries and their translations into
many languages. The UMLS is a formal ontology, but is not an ontology of the
semantic web. To increase the interoperability of this vocabulary towards a use by the
agents of the web, it is necessary to express the UMLS in a semantic web language
such as OWL.
Goal: The goal of this research is to translate the entire UMLS’s Semantic Network
and Metathesarus in a relational representational model towards a representation form
of OWL 2. To do this, we are relying on the hypothesis that the translation is possible
since the UMLS is structured with an ontological architecture.
Related work. Many past works [
2-5
] suggested methods of translation of the UMLS
in ontologies of the semantic web by addressing the formalization of the Semantic
Network of the UMLS either so as to emphasize the semi-formal character of the
UMLS [
2
], either to use the description logic (DL) so as to reveal inconsistencies
between statements originating from different data sources [
3
] or to evaluate the
capacity of an OWL-UMLS to interconnect different agents to UMLS [
4
], and finally,
to formalize the semantic network module of the UMLS so that it serves as a basic
vocabulary to an upper-ontology [
5
].
1 Methodology
Procedure: Fig. 1 presents the process divided in three steps, from the transformation
of the semantic network and of the ontological UMLS Metathesarus in text format to
the OWL 2 syntax. The first step consists in using MetamorphoSys, the installation
wizard suggested by UMLS, to convert the text format UMLS to a UMLS in a
MySQL Database. The second step aims to produce an UMLS-OWL by the
translation of the UMLS relational representation to an OWL representation according to the
architecture of the mapping in Fig. 2 with a combination of Java Persistence API and
Apache Jena OWL-API1 technologies. Third and last step consists in storing the
UMLS-OWL in a triplestore and to making it available on the web through a Linked
Open Data (LOD)2 server so that it can be accessed by a SPARQL-endpoint.
UMLS primitive classification in the DL architecture: Fig. 2 introduces the
repartition of the UMLS primitives in the architecture of the DL architecture who they
contain four levels of abstraction:
1. Metadata level: it encompasses all the primitives associated with all table
names and columns of the different tables of the UMLS
1 https://jena.apache.org/
2 Blazegraph (https://www.blazegraph.com/) in our case
2. Semantic level: it includes the TUI associated to the typological referencing
3. Conceptual level: it consists of the conceptual identities (CUI)
4. Factual level: it contains the different atomic entities (AUI) and roles (RUI)
Each primitive of the UMLS is classified according to the architecture of the DL that
is divided in 2 abstraction levels: the DL abstract level that includes the T-BOX and
R-BOX primitives and the DL factual level that stores the elements of the A-BOX.
The experimentation that was performed consisted in validating the usage of the
UMLS-OWL as a fitting vocabulary for the semantic use of a database. It was also
performed to validate the diffusion of the database on a LOD server. Fig. 3 presents
an overview of the programed scenario usin g the UMLS-OWL vocabulary. In the left
column, a taxonomic view the UMLS semantic networking is presented. In the right
part of Fig. 3, in G-OWL [
6
] notation, is presenting the association has_manifestion
of 5 different illnesses (names in bold) with the Findings and Signs or Symptoms .
In order to demonstrate the interoperability of the UMLS-OWL, the above scenario
has been put online in the LOD 3 server and interrogated by RelFinder[
7
]4 (see Fig 4).
3 http://iswc2017.plexusld.com:8080/bigdata/#query
4 http://iswc2017.plexusld.com/relfinder