BioMed Xplorer

Exploring (Bio)Medical Knowledge using Linked Data

Mohammad Shafahi, Hayo Bart and Hamideh Afsarmanesh

Informatics Institute, Faculty of Science, University of Amsterdam, Science Park 904, Amsterdam, The Netherlands

Keywords:

BioMed Xplorer, Disease Related Information, Semantic Web, Knowledge Base Ontology, Visualization,

Provenance Data, Medical Knowledge, External Data Source, RDF, Graph, Knowledge Exploration.

Abstract:

Developing an effective model for predicting risks of a disease requires exploration of a vast body of

(bio)medical knowledge. Furthermore, the continuous growth of this body of knowledge poses extra chal-

lenges. Numerous research has attempted to address these issues through developing a variety of approaches

and support tools. Most of these tools however, do not sufﬁciently address the needed dynamism, lack in-

tuitiveness in their use, and present a rather scarce amount of information usually obtained from a single

source. This research aims to address the aforementioned gaps through the development of a dynamic model

for (bio)medical knowledge, represented as a network of interrelated (bio)medical concepts, and integrating

disperse sources. To this end, this paper introduces BioMed Xplorer, presenting a model and a tool that enables

researchers to explore biomedical knowledge, organized in an information graph, through a user friendly and

intuitive interface. Furthermore, BioMed Xplorer provides concept related information from a multitude of

sources, while also preserving and presenting their provenance data. For this purpose a RDF knowledge base

has been created based on a core ontology which we have introduced. Results are further experimented with

and validated by some domain experts and are contrasted against the state of the art.

1 INTRODUCTION AND

RESEARCH APPROACH

The (bio)medical ﬁeld is vast and dynamic, with

knowledge developing rapidly as a result of contin-

uously ongoing research. Within this ﬁeld, extensive

research is conducted into identifying risk factors of

diseases as well as assessing their effect on the pres-

ence and associated severity of a disease. The avail-

able knowledge from this research on risk factors en-

ables researchers to develop models for risk predic-

tion, which might be used by practitioners to assess

someone’s risk on developing a particular disease.

Conventional methods for developing such models for

risk prediction would involve identifying the risk fac-

tors and their effects from the ever-evolving body of

(bio)medical knowledge. Achieving this aim would

thus involve checking vast amount of scientiﬁc publi-

cations for relevant statements regarding factors that

might affect a disease. This, however, is a cumber-

some and costly activity, especially when considering

the fact that the U.S. National Library of Medicine’s

(NLM) bibliographic database MEDLINE, as of to-

day, contains over 22 million citations, over 750,000

of which were added in 2014 (U.S. National Library

of Medicine, 2015b), and that these numbers have

grown exponentially (Hunter and Cohen, 2006). As

a result of the sheer size and continuous growth of

the body of (bio)medical knowledge, exploration of

this body of knowledge, as well as ﬁnding the relevant

knowledge for inclusion in models for risk prediction,

becomes increasingly challenging for researchers, po-

tentially causing an information overload (Hunter and

Cohen, 2006; Lu, 2011).

Numerous researchers have reckoned this prob-

lem and have attempted to address it from differ-

ent perspectives (Lu, 2011; Cohen and Hersh, 2005),

for example through the development of comprehen-

sive visualizations that represent knowledge extracted

from (bio)medical publications (Plake et al., 2006;

Rebholz-Schuhmann et al., 2007; Tao et al., 2005;

Kilicoglu et al., 2008; Bodenreider, 2000).

Even though the visual nature of these knowledge

representation and visualization tools provides them

with great expressive power, four common shortcom-

ings can be identiﬁed among them, being: i) their re-

stricted scope, focusing just on a particular sub-do-

main of the (bio)medical ﬁeld, ii) their lack of in-

Shafahi, M., Bart, H. and Afsarmanesh, H.

BioMed Xplorer - Exploring (Bio)Medical Knowledge using Linked Data.

DOI: 10.5220/0005700300510062

In Proceedings of the 9th International Joint Conference on Biomedical Engineering Systems and Technologies (BIOSTEC 2016) - Volume 3: BIOINFORMATICS, pages 51-62

ISBN: 978-989-758-170-0

tuitiveness and rather sharp learning curve, iii) the

scarce amount of information represented, solely lim-

ited to names and identiﬁers, lacking descriptions

or deﬁnitions, while these are available externally,

and iv) the fact that they are either no longer ac-

tive (AliBaba, PGviewer), or do not work properly

(EBIMed). From these shortcomings it thus becomes

clear that there is a need for a meaningful represen-

tation of the available (bio)medical knowledge that:

a) is intuitive, and b) represents information from

multiple sources. As such the following research

question can be conceived:

Can we develop a model of the (bio)medical

knowledge that is available from large, disperse, het-

erogeneous, and dynamic sources across the web?

In order to address this research question a ﬁve-

phase research approach has been designed, consist-

ing of the following phases: 1) State of the Art As-

sessment, 2) Data Source Characterization and Se-

lection, 3) Data Preprocessing and Ontology Design,

4) Data Interlinking and Fusion with external sources,

and 5) Model Visualization.

Completion of these ﬁve phases delivers a system

with an architecture that is shown in Figure 1. As

one might notice, the architecture consists of four core

modules, each of which corresponds to one of the ma-

jor design and development stages. The components

of these modules will be gradually deﬁned in the cor-

responding sections, as such fully describing the sys-

tem architecture.

The remainder of this paper is structured accord-

ing to the ﬁve phases that were outlined above, with

each section elaborately discussing a particular phase

of the research. In section 2, the characterization and

selection of data sources for inclusion in the model

is described. This is followed by a discussion of the

data preprocessing and ontology design in section 3,

whereas section 4 covers the fusion and interlinking

of the data. The visualization of the model is subse-

quently discussed in section 5, while the work is val-

idated in section 6. Finally, section 7 concludes the

paper.

2 DATA SOURCE

CHARACTERIZATION AND

SELECTION

Central to the development of a model is the data that

eventually will be represented in the model and thus

needs to be utilized for building and populating the

model. With the research question in mind, the iden-

tiﬁcation, and subsequent selection, of data sources

that provide disease related information, pertaining

to, for example, symptoms, inheritability, and genet-

ics of a disease, thus are the ﬁrst key steps in the

development process of the disease related informa-

tion model. A search for disease related information

results in a wide variety of structured (i.e standard-

ized terminologies or vocabularies, ontologies, and

databases) and unstructured (e.g. websites (U.S. Na-

tional Library of Medicine, 2015c; WebMD, LLC,

2015)) data sources. Data from unstructured sources

requires conversion to a structured format, for exam-

ple using Natural Language Processing (NLP) tech-

niques, and thus cannot be directly incorporated into

the disease related information model. As a result

we have decided to only incorporate structured data

sources. It is essential to designate a primary data

source for the development of the disease model as the

available structured data sources for disease related

information overlap in terms of covering the same in-

formation in different formats and presentations.

A disease model that is represented in a network-

like format consists of two components, namely con-

cepts and relationships among these concepts. Con-

cepts can be sourced from standardized terminolo-

gies, or from ontologies. Some well-known termi-

nologies in the biomedical ﬁeld are the International

Classiﬁcation of Diseases (ICD) (World Health Orga-

nization, 2015), Medical Subjects Headings (MeSH)

(U.S. National Library of Medicine, 2015a), and Sys-

tematized Nomenclature of Medicine Clinical Terms

(SNOMED CT) (International Health Terminology

Standards Development Organisation, 2015), whereas

the National Cancer Institute Thesaurus (NCIt) (U.S.

National Cancer Institute, 2015b), the Disease On-

tology (Institute for Genome Sciences - University

of Maryland School of Medicine, 2015), and the

Gene Ontology (Ashburner et al., 2000) are among

the frequently used ontologies within the biomedi-

cal ﬁeld. Instead of sourcing concepts from one or

multiple individual terminologies, one can source the

concepts from the Uniﬁed Medical Language Sys-

tem (UMLS) Metathesaurus (U.S. National Library

of Medicine, 2015e) or the National Cancer Institute

Metathesaurus (NCIm) (U.S. National Cancer Insti-

tute, 2015a), both of which integrate, among many

others, the aforementioned sources into a single ter-

minology. Using these metathesauri provides the op-

portunity of broadening the scope of the concepts that

are covered and, as such, expanding the knowledge

base of the model by using concepts represented in

the majority of separate terminologies. Therefore, the

use of either the UMLS or NCIm to deﬁne concepts

in the model is preferred over the use of separate ter-

minologies.

BIOINFORMATICS 2016 - 7th International Conference on Bioinformatics Models, Methods and Algorithms

Figure 1: System Architecture of BioMed Xplorer.

Relationships, on the other hand, can also be

sourced from the UMLS and NCIm. More extensive

relationships, however, can be obtained from the On-

line Mendelian Inheritance in Man (OMIM) database

(Johns Hopkins University, 2015), MalaCards (Weiz-

mann Institute of Science, 2015), or SemMedDB

(Kilicoglu et al., 2012). Considering the overarch-

ing aim of this research in aiding (bio)medical re-

searchers in their knowledge explorations efforts, and

due to the fact that (bio)medical knowledge originat-

ing from peer-reviewed literature is considered trust-

worthy and rich, relationships directly derived from

(bio)medical literature are selected as the primary re-

lationships in the model. To this end, SemMedDB is

thus selected as the primary source, presenting dis-

ease related information, for incorporation into the

developed model. This choice is further motivated

by the fact that SemMedDB is considerably larger

(containing over 70 million statements) than the other

identiﬁed sources containing disease related informa-

tion. Finally, the broad scope, covering terms across

the entire biomedical domain, also played a role in the

choice for SemMedDB.

3 DATA PREPROCESSING

Due to the large amounts of heterogeneous and dy-

namic information that is nowadays available across

a multitude of sources, relational databases are con-

sidered to be less than ideal for storing and instanti-

ating knowledge representations of information with

such nature (Hendler, 2014). Linked data, on the

other hand, provides a promising solution to this issue

as it is able to cope with such large amounts of dy-

namic and heterogeneous information (Berners-Lee

et al., 2001). To this end we therefore aim to de-

velop our model using Semantic Web technologies.

According to (Berners-Lee et al., 2001; Antoniou and

Van Harmelen, 2004) the Semantic Web consists of

three main components, being i) labeled graphs that

encode meaning by representing concepts and the re-

lations among them, and are usually expressed as

(subject-predicate-object) triples in RDF, ii) Uniform

Resource Identiﬁers (URIs) to uniquely identify the

items in the datasets as well as to assert meaning,

which is reﬂected in the design of RDF, and iii) on-

tologies to formally deﬁne the relations that can ex-

ist among data items. In order to develop our model

using the Semantic Web, the existence of these three

components needs to be ensured. Processing the data

in SemMedDB such that these three components ex-

ist, is therefore the main aim of the preprocessing

stage.

3.1 Ontology Design

In order to be able to generate labeled graphs from a

relational database, such as SemMedDB, and to en-

sure the use of URIs, an ontology needs to be devel-

oped that represents the desired data structure of these

graphs. This ontology should deﬁne the data items,

as well as the relations among them, that are aimed to

be represented. Considering that the planned model

should represent the statements, and their provenance

data, in SemMedDB as a RDF graph, it is key for the

ontology to closely resemble SemMedDB’s database

design. Prior work has been conducted in this area by

(Tao et al., 2012). In their work, (Tao et al., 2012)

aimed to optimize the organization and representa-

tion of Semantic MEDLINE data (SemMedDB) for

translational science studies by reducing redundancy

BioMed Xplorer - Exploring (Bio)Medical Knowledge using Linked Data

through the application of Semantic Web technolo-

gies. This is achieved by representing the concepts

and associations in SemMedDB as RDF. Despite suc-

cessfully decreasing the redundancy of the informa-

tion in SemMedDB, two shortcomings can be identi-

ﬁed in the ontology that was developed by (Tao et al.,

2012). First of all, the ontology represents a limited

amount of information compared to the information

that is available in SemMedDB. This, in turn, impedes

the ability to incorporate external resources into the

model since among the information from SemMedDB

that is omitted are unique identiﬁers that are required

to retrieve the appropriate entities from these exter-

nal sources. The second shortcoming is the lack of

reuse of terms deﬁned in existing vocabularies, which

is one of the founding principles of the Semantic Web

(Shadbolt et al., 2006). In (Tao et al., 2012), the devel-

oped ontology deﬁnes all terms used, whereas equiv-

alent classes might already exist in other vocabularies

in the Web of Data. Such reuse would facilitate the

linking of data to a Web of Data, which is an overar-

ching goal of the Semantic Web (Berners-Lee et al.,

2001).

Despite the limitations of the ontology developed

in (Tao et al., 2012), this ontology is considered as a

starting point as well as an opportunity to improve on

and extend upon. To this extent, the BioMed Xplorer

Ontology is developed that addresses the identiﬁed

shortcomings by representing most of the information

contained within SemMedDB, as well as by reusing

as much terms from existing vocabularies or ontolo-

gies as possible. The BioMed Xplorer Ontology is de-

veloped in the Web Ontology Language (OWL2) and

is published on a Persistent Uniform Resource Loca-

tor (PURL) (Weibel et al., 1996) domain

. Such a lo-

cator allows the underlying Web address of a resource

to change while not affecting the availability of the

systems that depend on this resource. The BioMed

Xplorer Ontology is shown in Figure 2.

RDF Reiﬁcation. Considering that the provenance

data in SemMedDB applies to statements as a whole,

reiﬁcation is necessary in order to represent this

provenance data in the ontology. (Tao et al., 2012)

also recognized this need, however, they did not use

the RDF Reiﬁcation vocabulary as outlined in (World

Wide Web Consortium et al., 2014). The BioMed

Xplorer Ontology on the other hand implements the

RDF reiﬁcation vocabulary.

As the statements contained in SemMedDB relate

two UMLS concepts to each other, both the subject

and object of an rdf:Statement instance are modelled

http://purl.org/net/fcnmed

as instances of a Concept class. The concepts are

related to each other through one, of 58, relation-

ships that are identiﬁed by SemRep (U.S. National

Library of Medicine, 2015d). The predicate of an

rdf:Statement instance therefore is modelled as one

of 58 instances of the Relation class. This set of re-

lationships consists of two disjunctive subsets, with

one subset containing 31 relationships derived from

the UMLS Semantic Network, such as ”causes”, and

the other subset containing the remaining 27 relation-

ships, which are negated versions of the relationships

in the ﬁrst subset, such as ”neg causes”, referring

to ”does not causes” (Kilicoglu et al., 2012). Rela-

tionships belonging to the negated subset are preﬁxed

with ”NEG”, whereas all other relationships are con-

sidered to belong to the subset of afﬁrmed relation-

ships. These two subsets of relations are represented

in the BioMed Xplorer Ontology as two subclasses of

the Relation class, being the AfﬁrmedRelation and the

NegatedRelation classes respectively.

The provenance data in SemMedDB includes both

the sentences from which a statement is derived, as

well as the publications in which these sentences oc-

cur. Reiﬁcation of the statements enables the asser-

tion of this provenance data to their respective state-

ments. To this end, sentences are represented as in-

stances of the Sentence class, which are related to the

rdf:Statement class through a derivedFrom property.

The articles in which these statements and sentences

are contained, are represented as instances of an Arti-

cles class, which are related to the rdf:Statement class

through a source property. Furthermore, sentences are

related to articles through the partOf property, indi-

cating that a sentence is part of an academic article.

In addition to the object properties, relating classes

to each other, discussed in this section, a number

of datatype properties, associating data values (such

as identiﬁers) to classes, are asserted to each of the

classes in the BioMed Xplorer Ontology as well. Col-

lectively, these properties aim to represent as much

information from SemMedDB in the ontology as pos-

sible.

Vocabulary Reuse. The BioMed Xplorer Ontology

aims to reuse as much existing classes and proper-

ties as possible. To this extent all elements of the

ontology, which include the classes and both the ob-

ject and datatype properties, except elements from the

RDF or RDFS namespaces, have been checked for

the presence of already deﬁned equivalent concepts

or properties in existing vocabularies. This has been

accomplished by making use of the online RDF vo-

cabulary search and lookup tool vocab.cc (Institute

of Applied Informatics and Formal Description Meth-

BIOINFORMATICS 2016 - 7th International Conference on Bioinformatics Models, Methods and Algorithms

Figure 2: BioMed Xplorer Ontology, only showing the object properties.

ods - Karlsruhe Research Institute, 2015) that allows

one to enter any term, and returns any classes and

properties that (partially) match the term. In gen-

eral the highest ranked term that corresponds to the

role of the term in the BioMed Xplorer Ontology

(e.g. class or property) is selected for reuse in the

BioMed Xplorer. In the end, the search for existing

terms lead to the incorporation of terms from three

existing vocabularies, being i) the Bibliographic On-

tology (D’Arcus and Giasson, 2015), ii) the Dublin

Core Metadata Terms (Dublin Core Metadata Initia-

tive (DCMI), 2015), and iii) the Simple Knowledge

Organization System (World Wide Web Consortium,

2015).

4 DATA INTERLINKING &

FUSION

The ontology developed in section 3.1 deﬁnes the de-

sired data structure for the developed model. Generat-

ing the labeled graphs from the SQL in SemMedDB,

however, requires a mapping that speciﬁes how the

data in the database is matched and converted to

the appropriate class instances, properties, and prop-

erty values speciﬁed in the ontology. Such a map-

ping can be developed using D2RQ, a declarative

language for describing mappings between relational

databases, RDF(S), and OWL ontologies (Bizer and

Seaborne, 2004). The developed D2RQ mapping ﬁles

have been made available online

. A mapping ﬁle en-

ables RDF applications to access relational databases

as virtual RDF graphs through the companion tool

D2R Server (Bizer and Cyganiak, 2006). These vir-

tual RDF graphs can subsequently be queried using

The mapping ﬁles are available online from:

https://goo.gl/1yD0WO

the SPARQL protocol, with the D2RQ mapping trans-

lating the SPARQL queries to SQL queries, and trans-

lating the query results back to RDF. Both D2RQ

and D2R are jointly available in the D2RQ Platform

(Bizer and Seaborne, 2004). With the developed map-

ping ﬁle, the data in SemMedDB can be interlinked

as RDF triples according to the speciﬁed ontology, as

such surfacing and populating the actual disease re-

lated information model. Furthermore, the combina-

tion of the ontology and the use of RDF ensures the

ability to link to the data in the information model

from external datasets, through the URIs assigned to

instances and properties.

In order to achieve complete data fusion with ex-

ternal sources, as such creating a truly Linked Data

model, the data in the disease related information

model should be linked to related entities or instances

in external (RDF) data sources. This can be achieved

by setting RDF links between the data in the model

and these external data sources (Berners-Lee et al.,

2009). One common way of setting such links be-

tween data sets is through the owl:sameAs property,

which indicates that two linked individuals refer to

the same thing (Dean et al., 2004). Establishing these

links subsequently enables the incorporation of data

from the external data sources into the disease related

information model. Key to this data fusion process is

the identiﬁcation of external data sources containing

instances that are equivalent to the instances in the

developed model. The search for these data sources

containing equivalent instances has been facilitated

by searching the Linked Open Data cloud

for unique

standardized instance identiﬁers. Among these iden-

tiﬁers in SemMedDB are the UMLS Concept Unique

Identiﬁer (CUI), the Entrez-Gene ID, and the OMIM

identiﬁer for concepts, as well as the PubMed Iden-

tiﬁer (PMID) for publications. The search of the

For details see http://lod-cloud.net/

BioMed Xplorer - Exploring (Bio)Medical Knowledge using Linked Data

Linked Open Data cloud for (bio)medical RDF data

sources that represent either (bio)medical concepts,

identiﬁed by one of the aforementioned identiﬁers,

or publications, identiﬁed by the PubMed identiﬁer,

returned two main external data sources that could

be fused with the data in SemMedDB: Linked Life

Data (Momtchev et al., 2009), and Bio2RDF (Belleau

et al., 2008).

5 BIOMED XPLORER UI

Assisting (bio)medical researchers in their knowledge

exploration efforts can be achieved by enabling them

to intuitively explore the body of (bio)medical knowl-

edge. To this end it is therefore imperative to visualize

the developed disease related information graph, rep-

resenting this body of knowledge, that incorporates

other disease related information gathered and aggre-

gated from disperse sources across the Web. With

this in mind three key requirements for the BioMed

Xplorer UI can be imagined, being that it should:

i) be usable and intuitive (e.g. supported by an ap-

propriate visualization paradigm), ii) concisely repre-

sent provenance data (e.g. the publications as well

as sentences from which statements are derived), and

iii) represent information from multiple sources (e.g

concept summaries and deﬁnitions). Based on these

identiﬁed requirements, BioMed Xplorer has been

developed and made available

on the Web. The

BioMed Xplorer UI supports the visualization of the

information and has been developed in JavaScript in

combination with the d3.js

and jQuery

libraries.

The data visualized by the BioMed Xplorer UI

is obtained from BioMed Xplorer’s back-end, which

consists of a Virtuoso triple store containing the dis-

ease related information model in RDF. This triple

store provides a built-in SPARQL endpoint that can be

queried by BioMed Xplorer using the SPARQL (Har-

ris et al., 2013) protocol. Efﬁcient query handling has

been achieved by the development of a caching mech-

anism.

The developed user interface has three key fea-

tures being: i) a graph-based visualization, ii) the ex-

ploration of concept information, and iii) the explo-

ration and assessment of relationships. Each of these

three features will be brieﬂy discussed in the remain-

der of this section.

BioMed Xplorer is available http://goo.gl/qeuW5k

(best viewed in Firefox).

For details see http://d3js.org/

For details see http://jquery.com/

Graph Visualization. The BioMed Xplorer UI em-

ploys a graph-based visualization of (bio)medical

knowledge as shown in Figure 3. Exploration and

traversal of the knowledge graph is supported through

the expansion of concepts (by double clicking on con-

cepts) and collapsing of concepts (by right clicking on

concepts). Additionally, panning (by click and drag)

and zooming (by scrolling) is supported as well.

Exploring Concept Information. Within the

BioMed Xplorer UI concept information can be

explored through concept summaries, which can

be opened by clicking on concepts, and concept

overviews (as shown in Figure 4), which can be

opened by choosing to show details in a concept

summary. Concept information includes a wide range

of information available from within the model as

well as from external sources, such as Linked Life

Data and Bio2RDF.

Exploring and Assessing Relationships. Relation-

ships between concepts can be explored in the

BioMed Xplorer UI through statement summaries,

which can be opened by clicking on an edge, and

statement overview, which can be opened by choos-

ing to show details in a relationship summary. Within

statement overviews, a wide range of statement infor-

mation is available, as is shown in Figure 5. Among

the available information is: the complete statement,

two aggregates of the available provenance data, a

brief overview of the source and target concepts of

the relationship, the sentences from which the state-

ment is derived at a publication level, as well as the

details of the publications.

6 VALIDATION

Keeping the key roles of both the BioMed Xplorer

Ontology, as the the foundation for the knowledge

base, and the BioMed Xplorer UI, as the visualization

of this knowledge base in mind, the validation of these

two outcomes is imperative. This validation aims

to assess whether the proposed solutions successfully

address the already identiﬁed gap as well as how the

proposed solutions measure against existing work. To

this end, a two folded validation of both the ontology

and the visualization has been conducted. In this re-

gard a comparison to prior work has been conducted

ﬁrst, the results of which are shown in Tables 1 and

2 respectively. Secondly, an evaluation has been per-

formed by 6 experts in the ﬁeld. Results of this expert

evaluation showed that both the BioMed Xplorer On-

tology and the BioMed Xplorer UI successfully sat-

BIOINFORMATICS 2016 - 7th International Conference on Bioinformatics Models, Methods and Algorithms

Figure 3: A screenshot from BioMed Xplorer UI and its graph-based visualization of (bio)medical knowledge, representing

(bio)medical concepts as nodes and their interrelationships as edges.

Figure 4: BioMed Xplorer UI’s Concept Overview for ”Malignant Neoplasms”.

isfy the identiﬁed requirements, with average grades

of a 7.8 and 7.6 out of 10 respectively. Details of the

expert evaluation of both the BioMed Xplorer Ontol-

ogy and the BioMed Xplorer UI are provided in Ta-

bles 3 and 4.

Comparison to Related Work. There are ﬁve main

knowledge representation and visualization tools

identiﬁed that attempt to address similar challenges

associated with exploring the body of (bio)medical

knowledge through the representation and visualiza-

tion of the knowledge contained within scientiﬁc pub-

lications. Among these tools are: AliBaba (Plake

et al., 2006), EBIMed (Rebholz-Schuhmann et al.,

2007), PGviewer (Tao et al., 2005), Semantic MED-

LINE (Kilicoglu et al., 2008), and the Semantic Nav-

igator (Bodenreider, 2000). Due to the close corre-

spondence between the aims of these tools and the

aims of our research, these ﬁve tools are considered

as the base for comparison to BioMed Xplorer. A

BioMed Xplorer - Exploring (Bio)Medical Knowledge using Linked Data

Figure 5: BioMed Xplorer UI’s Statement Overview for ”Malignant Neoplasms” part of ”Rattus Norvegicus”.

Table 2: Comparison of BioMed Xplorer UI with ﬁve knowledge visualization tools developed in prior work.

Characteristic AliBaba EBIMed PG-

viewer

Semantic

MEDLINE

Semantic

Navigator

BioMed

Xplorer

Scope Limited Limited Limited Biomedical Biomedical Biomedical

Available No No No Yes Yes Yes

Visualization paradigm Graph Tabular Tree Graph Graph Graph

Concept categorization Yes Yes Yes Yes No Yes

Incorporation of links to external

sources

Yes Yes No Yes No Yes

Incorporation of data from external

sources

Yes Yes Yes No No Yes

Presentation of concept related infor-

mation

Yes No Yes Yes No Yes

Incorporation of provenance data Yes Yes Yes Yes No Yes

Table 1: Comparison of BioMed Xplorer Ontology with the

ontology developed by(Tao et al., 2012)

Characteristic Tao et al.,

2012

BioMed

Xplorer

Ontology

RDF Reiﬁcation No Yes

Vocabulary reuse No Yes

Links to external

data sources

No Yes

Number of related

data-items captured

4 17

Provenance data cap-

tured

Publications Publications

and sentences

comparison of the characteristics of these ﬁve selected

tools is provided in Table 2.

AliBaba acts as an interactive tool that graphi-

cally summarizes the associations between concepts

from a rather limited sub-domain of the (bio)medical

ﬁeld, namely between cells, diseases, drugs, proteins,

species, and tissues. AliBaba extracts these concepts

and the associations between them from scientiﬁc

publications that match a PubMed query.

Semantic MEDLINE provides similar functional-

ity in a broader domain as it uses concepts in the

UMLS Metathesaurus as its base. These concepts,

and their relationships, are extracted, respectively

identiﬁed, from the complete MEDLINE database,

and, similar to AliBaba, subsequently presented as a

graph. The Semantic Navigator also employs a graph-

based format. In this tool, the graph is used to repre-

sent the semantic structure of the UMLS, and as such

enables users to visually explore the concepts in the

UMLS as well as their relationships.

Contrary to the graph-based format employed by

AliBaba, Semantic MEDLINE, the Semantic Naviga-

tor, and BioMed Xplorer for visualizing (bio)medical

BIOINFORMATICS 2016 - 7th International Conference on Bioinformatics Models, Methods and Algorithms

Table 3: Frequency distribution of the ﬁve point Likert-

scale scores for evaluating the BioMed Xplorer Ontology.

A score of 1 indicates disagreement and 5 indicates agree-

ment.

Statement 1 2 3 4 5

The ontology is capable of

representing (statements of)

biomedical knowledge

0 0 1 4 1

The ontology models (state-

ments of) biomedical knowl-

edge appropriately

0 1 1 3 1

The ontology is capable of rep-

resenting the provenance data

associated with (statements of)

biomedical knowledge

0 0 2 2 2

The ontology models prove-

nance data associated with

(statements of) biomedical

knowledge appropriately

0 1 1 2 2

The ontology globally ﬁts its

purpose

0 0 2 2 2

Table 4: Frequency distribution of the ﬁve point Likert-

scale scores for evaluating the BioMed Xplorer UI. A score

of 1 indicates disagreement and 5 indicates agreement.

Statement 1 2 3 4 5

The implemented function-

alities support and facilitate

the exploration of biomedical

knowledge

0 1 1 1 3

Color coding of the nodes is

helpful

0 0 1 2 3

The information in the sum-

maries has a clear structure

0 1 2 1 2

The information in the sum-

maries is relevant

0 1 1 3 1

The information in the ex-

tended details has a clear struc-

ture

0 1 1 3 1

The information in the ex-

tended details is relevant

0 0 1 2 3

The interface is well structured

/ organized

0 0 2 3 1

The graphical user interface

has an adequate look and feel

0 0 2 2 2

The tool behaves as expected 0 2 2 1 1

The visualization is intuitive in

its use

0 2 1 2 1

The visualization of informa-

tion is simple and smooth

0 1 0 3 2

The system globally ﬁts its pur-

pose

0 2 0 1 3

knowledge, EBIMed and PGviewer make use of

two alternative visualization paradigms. On the one

hand, EBIMed identiﬁes relationships between a set

of (bio)medical concepts extracted from publications

that match a MEDLINE query, and visualizes these

in a tabular format. The concepts represented in

EBIMed stem from the (bio)medical subdomain con-

sisting of proteins, Gene ontology annotations, drugs,

and species. On the other hand, PGviewer employs

tree visualization, with the purpose of clustering, in

order to present relationships from the genotype and

phenotype subdomain that are stored both in struc-

tured, such as MEDLINE, and (unstructured) textual

databases, such as the OMIM.

The ﬁve tools identiﬁed from prior work, in sum-

mary, thus have two main shortcomings. On the

one hand they focus on a particular subdomain of

the (bio)medical ﬁeld (AliBaba, EBIMed, PGviewer)

and, as such, inhibit the exploration of the body

of (bio)medical knowledge. On the other hand

they employ an alternative visualization paradigm

(EBIMed and PGviewer) that is less focused on the

visual representation of knowledge. The BioMed

Xplorer UI overcomes these shortcomings, as such

improving over most tools developed in prior work,

through its broad scope, aiming to cover the complete

(bio)medical domain, and its graph-based visualiza-

tion. More speciﬁcally, BioMed Xplorer can be con-

sidered to be on par with Semantic MEDLINE and

the Semantic Navigator as these two tools both fo-

cus on the entire (bio)medical ﬁeld as well as employ

a graph-based paradigm for visualizing (bio)medical

knowledge. The three aforementioned tools are fur-

thermore available on the Web, whereas AliBaba,

EBIMed, and PGviewer are no longer available. The

position of BioMed Xplorer is further reinforced by

the fact that it is the only tool that is based on RDF,

which improves its ability to handle large amounts

of heterogeneous data from disperse sources. Other

tools, on the other hand, are based on traditional re-

lational databases, as such inhibiting their ability to

incorporate data from additional external sources into

these tools.

In addition to representing (bio)medical knowl-

edge through statements that relate two (bio)medical

concepts to each other, the presentation of concepts

and statements related information is also of great im-

portance, as it provides background knowledge about

the concepts involved in the statements or about the

statements themselves. To this end, BioMed Xplorer

is on par with all of the other tools considering the

presentation of statement related information. This

information typically includes the complete statement

itself, including its source and target concepts, the

type of the statement, as well as the provenance data

associated with the statements in terms of the abstract

or sentences, and publications from which the state-

ments were derived. Such provenance data is pro-

vided by all the tools included in the comparison, ex-

BioMed Xplorer - Exploring (Bio)Medical Knowledge using Linked Data

cept for the Semantic Navigator as this tool solely rep-

resents the relationships stored in the UMLS. Occa-

sionally, the statement related information might also

include aggregates of the provenance data, such as the

number of sentences and publications from which a

particular statement is derived, as is the case for Se-

mantic MEDLINE and BioMed Xplorer. Whereas the

BioMed Xplorer is on par with the other tools in re-

lation to the presentation of statement related infor-

mation and the incorporation of provenance data, it

in fact improves over these tools on the presentation

of concepts related information. The concepts related

information presented in BioMed Xplorer UI extends

well beyond the conventional information that is in-

corporated. While, tools such as EBIMed and the Se-

mantic Navigator do not present any of such concepts

related information at all, data items such as (seman-

tic) types, synonyms, and parts of publications that

mention the particular concept are presented by Al-

iBaba, PGviewer, and Semantic MEDLINE. BioMed

Xplorer extends this further through the incorporation

of a wide range of cross-identiﬁers of concepts, a def-

inition, and a range of data items pertaining to the

clinical features, diagnosis, inheritance, pathogenesis,

and genetics of a disease from OMIM, if available.

The presentation of this wide range of concept related

information in BioMed Xplorer is partially facilitated

through the incorporation of data from external (struc-

tured) data sources, including Linked Life Data and

Bio2RDF, which demonstrates its superiority com-

pared to the other tools developed in prior work.

Among these other tools, the incorporation of infor-

mation from such external sources is either largely

absent (such as in Semantic MEDLINE and Seman-

tic Navigator), or limited to the inclusion of informa-

tion from PubMed (such as in AliBaba, EBIMed, and

PGviewer). Links to external data sources, usually

in the form of cross-references to standardized termi-

nologies, on the other hand, are commonly used by

the tools developed in prior work, with only PGviewer

and the Semantic Network lacking such cross refer-

ences.

For the validation of the ontology, the ontology

developed by (Tao et al., 2012) is considered as the

base to which our developed ontology is compared.

A comparison of the characteristics of the two ontolo-

gies is provided in Table 1. As is clear from this table,

the ontology developed in this research improves the

ontology developed by (Tao et al., 2012) on a number

of aspects, which will be further discussed below, as

such contributing to the validation of the ontology de-

veloped in this research. As was discussed in section

3.1, reiﬁcation has been applied in both ontologies to

allow triples to involve a particular (bio)medical state-

ment, as a whole, into another statement, and thus

enable meta-statements: statements about statements.

This can be achieved by treating a statement, relat-

ing two (bio)medical concepts to each other through

a relation, as a separate entity to which the subject,

the predicate, and the object of the original statement

are assigned using an object property. The ontology

developed by (Tao et al., 2012) performs this by mak-

ing use of the Association class in combination with

the hass name, has predicate, and haso name prop-

erties. BioMed Xplorer Ontology, on the other hand,

makes use of the ofﬁcial RDF reiﬁcation vocabulary

that uses the rdf:Statement class in combination with

the rdf:subject, rdf:predicate, and rdf:object proper-

ties. Additionally, the use of this ofﬁcial RDF reiﬁ-

cation vocabulary also contributes to the reuse of ex-

isting vocabularies, one of the key principles of Se-

mantic Web (Shadbolt et al., 2006). To further pro-

mote this base principle of the Semantic Web, the de-

veloped ontology, in addition to the use of the RDF

reiﬁcation vocabulary, makes extensive use of exist-

ing classes and properties from other vocabularies.

This is a considerable improvement over the ontology

developed by (Tao et al., 2012), as the reuse of exist-

ing vocabularies, aside from the RDF vocabulary, is

not present in their ontology.

Since the purpose of the developed model is to en-

able researchers to explore the body of (bio)medical

knowledge as well as its (disease) related information,

the amount of information captured by the ontology

is of great importance. To this extent, the ontology

developed by (Tao et al., 2012) can be considered as

rather limited due to the fact that there is no direct

evidence of the incorporation of any (disease) related

information beyond the three datatype properties as-

signing names to concepts and relations, as well as

identiﬁers to publications. The ontology developed in

our research improves on this point by associating 17

data-type properties to the ontology classes that can

be used to capture a wide range of (disease) related

information. This is further facilitated by the incor-

poration of RDF links to the equivalent resources in

external data sources, including Linked Life Data and

Bio2RDF, which contain a wealth of (disease) related

information. No such links are incorporated in the on-

tology developed by (Tao et al., 2012).

Finally, the developed ontology extends the ontol-

ogy developed by (Tao et al., 2012) by incorporat-

ing the sentences from which the represented state-

ments are derived, in addition to those publications

from which these sentences are a part, as a compo-

nent of the provenance data that is associated to the

statements. The incorporation of these sentences pro-

vides additional value to the disease related informa-

BIOINFORMATICS 2016 - 7th International Conference on Bioinformatics Models, Methods and Algorithms

tion model as it enables the presentation of the direct

source of a particular statement, as opposed to the pre-

sentation of solely the publication from which a state-

ment is derived.

7 CONCLUSION

Two common shortcomings among (bio)medical

knowledge discovery representation and visualization

tools are the scarcity of the information that is rep-

resented, usually coming from a single source, as

well as the lack of intuitiveness. To address this gap,

this research aimed at developing a dynamic model

representing (bio)medical knowledge, available from

disperse sources across the Web, as a network of

inter-related (bio)medical concepts, while incorporat-

ing Semantic Web technologies to deal with large

amounts of dynamic and heterogeneous information.

To achieve this goal, a ﬁve phase research ap-

proach has been followed, consisting of: 1) State of

the Art Assessment, 2) Data Source Characterization

and Selection, 3) Data Preprocessing and Ontology

Design, 4) Data Interlinking and Fusion with exter-

nal sources, and 5) Model Visualization. Comple-

tion of these phases resulted in the development of

the BioMed Xplorer Ontology, providing a founda-

tion of the knowledge base, and the BioMed Xplorer

UI, acting as a visualization of the knowledge base.

Future work will focus on implementing key in-

dicators, representing the importance of instances, to

more efﬁciently regulate which concepts and state-

ments are presented to the user. To this end, indica-

tors such as the degree of concepts, or number of sen-

tences or publications from which a statement is de-

rived, might be used. A second point of future work

will focus on extending BioMed Xplorer’s function-

ality with extensive ﬁltering options, as such enabling

the user to view important, or less important, concepts

and statements based on key indicators.

ACKNOWLEDGEMENTS

This work was carried out on the Dutch national e-

infrastructure with the support of SURF Foundation

We also like to thank the School of Medicine at Dem-

ocritus University of Trace for helping with some re-

quirements identiﬁcation and validation.

For details visit: https://www.surf.nl/en/services-and-

products/hpc-cloud/index.html

REFERENCES

Antoniou, G. and Van Harmelen, F. (2004). A semantic web

primer. MIT press.

Ashburner, M., Ball, C. A., Blake, J. A., Botstein, D.,

Butler, H., Cherry, J. M., Davis, A. P., Dolinski, K.,

Dwight, S. S., Eppig, J. T., et al. (2000). Gene ontol-

ogy: tool for the uniﬁcation of biology. Nature genet-

ics, 25(1):25–29.

Belleau, F., Nolin, M.-A., Tourigny, N., Rigault, P., and

Morissette, J. (2008). Bio2rdf: towards a mashup to

build bioinformatics knowledge systems. Journal of

biomedical informatics, 41(5):706–716.

Berners-Lee, T., Bizer, C., and Heath, T. (2009). Linked

data-the story so far. International Journal on Seman-

tic Web and Information Systems, 5(3):1–22.

Berners-Lee, T., Hendler, J., Lassila, O., et al. (2001). The

semantic web. Scientiﬁc american, 284(5):28–37.

Bizer, C. and Cyganiak, R. (2006). D2r server-publishing

relational databases on the semantic web. In Poster at

the 5th International Semantic Web Conference, pages

294–309.

Bizer, C. and Seaborne, A. (2004). D2rq-treating non-

rdf databases as virtual rdf graphs. In Proceed-

ings of the 3rd international semantic web conference

(ISWC2004), volume 2004. Citeseer Hiroshima.

Bodenreider, O. (2000). A semantic navigation tool for the

umls. In Proceedings of the AMIA Symposium, page

971. American Medical Informatics Association.

Cohen, A. M. and Hersh, W. R. (2005). A survey of current

work in biomedical text mining. Brieﬁngs in bioinfor-

matics, 6(1):57–71.

D’Arcus, B. and Giasson, F. (2015). Bibli-

ographic Ontology Speciﬁcation (BIBO).

http://bibliontology.com/speciﬁcation. [Online;

accessed August 28 2015].

Dean, M., Schreiber, G., Bechhofer, S., van Harmelen, F.,

Hendler, J., Horrocks, I., McGuinness, D. L., Patel-

Schneider, P. F., and Stein, L. A. (2004). Owl web

ontology language reference. W3C Recommendation

February, 10.

Dublin Core Metadata Initiative (DCMI) (2015). Dublin

Core (DC). http://dublincore.org/. [Online; accessed

August 28 2015].

Harris, S., Seaborne, A., and Prudhommeaux, E. (2013).

Sparql 1.1 query language. W3C Recommendation,

21.

Hendler, J. (2014). Data integration for heterogenous

datasets. Big data, 2(4):205–215.

Hunter, L. and Cohen, K. B. (2006). Biomedical language

processing: what’s beyond pubmed? Molecular cell,

21(5):589–594.

Institute for Genome Sciences - University of Maryland

School of Medicine (2015). Disease Ontology (DO).

http://disease-ontology.org/. [Online; accessed Au-

gust 28 2015].

Institute of Applied Informatics and Formal Description

Methods - Karlsruhe Research Institute (2015). Vo-

cab.cc. http://www.vocab.cc/. [Online; accessed Au-

gust 28 2015].

BioMed Xplorer - Exploring (Bio)Medical Knowledge using Linked Data

International Health Terminology Standards Development

Organisation (2015). Systematized Nomencla-

ture of Medicine Clinical Terms (SNOMED CT).

http://www.ihtsdo.org/snomed-ct. [Online; accessed

August 28 2015].

Johns Hopkins University (2015). Online Mendelian Inher-

itance in Man (OMIM). http://www.omim.org/. [On-

line; accessed August 28 2015].

Kilicoglu, H., Fiszman, M., Rodriguez, A., Shin, D., Rip-

ple, A., and Rindﬂesch, T. C. (2008). Semantic med-

line: a web application for managing the results of

pubmed searches. In Proceedings of the third interna-

tional symposium for semantic mining in biomedicine,

volume 2008, pages 69–76. Citeseer.

Kilicoglu, H., Shin, D., Fiszman, M., Rosemblat, G., and

Rindﬂesch, T. C. (2012). Semmeddb: a pubmed-

scale repository of biomedical semantic predications.

Bioinformatics, 28(23):3158–3160.

Lu, Z. (2011). Pubmed and beyond: a survey of web

tools for searching biomedical literature. Database,

2011:baq036.

Momtchev, V., Peychev, D., Primov, T., and Georgiev,

G. (2009). Expanding the pathway and interaction

knowledge in linked life data. Proc. of International

Semantic Web Challenge.

Plake, C., Schiemann, T., Pankalla, M., Hakenberg, J., and

Leser, U. (2006). Alibaba: Pubmed as a graph. Bioin-

formatics, 22(19):2444–2445.

Rebholz-Schuhmann, D., Kirsch, H., Arregui, M., Gaudan,

S., Riethoven, M., and Stoehr, P. (2007). Ebimedtext

crunching to gather facts for proteins from medline.

Bioinformatics, 23(2):e237–e244.

Shadbolt, N., Hall, W., and Berners-Lee, T. (2006). The

semantic web revisited. Intelligent Systems, IEEE,

21(3):96–101.

Tao, C., Zhang, Y., Jiang, G., Bouamrane, M.-M., and

Chute, C. G. (2012). Optimizing semantic medline

for translational science studies using semantic web

technologies. In Proceedings of the 2nd international

workshop on Managing interoperability and compleX-

ity in health systems, pages 53–58. ACM.

Tao, Y., Friedman, C., and Lussier, Y. A. (2005). Vi-

sualizing information across multidimensional post-

genomic structured and textual databases. Bioinfor-

matics, 21(8):1659–1667.

U.S. National Cancer Institute (2015a). NCI Metathesaurus

(NCIm). https://ncim.nci.nih.gov. [Online; accessed

August 28 2015].

U.S. National Cancer Institute (2015b). NCI Thesaurus

(NCIt). https://ncit.nci.nih.gov. [Online; accessed Au-

gust 28 2015].

U.S. National Library of Medicine (2015a).

Medical Subject Headings(MeSH).

http://www.nlm.nih.gov/mesh/meshhome.html.

[Online; accessed August 28 2015].

U.S. National Library of Medicine

(2015b). Medline fact sheet.

http://www.nlm.nih.gov/pubs/factsheets/medline.html.

[Online; accessed August 28 2015].

U.S. National Library of Medicine (2015c). Medline Plus.

http://www.nlm.nih.gov/medlineplus/. [Online; ac-

cessed August 28 2015].

U.S. National Library of Medicine (2015d). SemRep.

http://semrep.nlm.nih.gov. [Online; accessed August

28 2015].

U.S. National Library of Medicine (2015e). Uni-

ﬁed Medical Language System (UMLS).

http://www.nlm.nih.gov/research/umls/. [Online;

accessed August 28 2015].

WebMD, LLC (2015). WebMD. http://www.webmd.com/.

[Online; accessed August 28 2015].

Weibel, S. L., Jul, E., and Shafer, K. E. (1996). PURLs:

Persistent uniform resource locators. OCLC Online

Computer Library Center.

Weizmann Institute of Science (2015). MalaCards.

http://www.malacards.org/. [Online; accessed August

28 2015].

World Health Organization (2015). Interna-

tional Classiﬁcation of Diseases (ICD).

http://www.who.int/classiﬁcations/icd/en/. [On-

line; accessed August 28 2015].

World Wide Web Consortium (2015). Simple

Knowledge Organization System (SKOS).

http://www.w3.org/2004/02/skos/. [Online; ac-

cessed August 28 2015].

World Wide Web Consortium et al. (2014). Rdf 1.1 seman-

tics.

BIOINFORMATICS 2016 - 7th International Conference on Bioinformatics Models, Methods and Algorithms