=Paper=
{{Paper
|id=None
|storemode=property
|title=The Knowledge-driven Exploration of Integrated Biomedical Knowledge Sources Facilitates the Generation of New Hypotheses
|pdfUrl=https://ceur-ws.org/Vol-783/paper6.pdf
|volume=Vol-783
|dblpUrl=https://dblp.org/rec/conf/semweb/NguyenBMS11
}}
==The Knowledge-driven Exploration of Integrated Biomedical Knowledge Sources Facilitates the Generation of New Hypotheses==
https://ceur-ws.org/Vol-783/paper6.pdf
The knowledge-driven exploration of integrated
biomedical knowledge sources facilitates the
generation of new hypotheses
Vinh Nguyen1 , Olivier Bodenreider2 , Todd Minning3 , and Amit Sheth1
1
Kno.e.sis Center, Wright State University, Dayton, Ohio
2
National Library of Medicine, Bethesda, Maryland
3
Center for Tropical and Emerging Global Diseases, Univeristy of Georgia, Georgia
Abstract. Knowledge gained from the scientific literature can comple-
ment newly obtained experimental data in helping researchers under-
stand the pathological processes underlying diseases. However, unless the
scientific literature and experimental data are semantically integrated, it
is generally difficult for scientists to exploit the two sources effectively.
We argue that, in addition to the semantic integration of heterogeneous
knowledge sources, the usability of the integrated resource by scientists
is dependent upon the availability of knowledge visualization and ex-
ploration tools. Moreover, the integration techniques must be scalable
and the exploration interfaces must be easy to use by bench scientists.
The end goal of such integrated knowledge sources and exploration tools
is to enable scientists to generate novel hypotheses from the knowledge
they explore. We tested the feasibility of our approach on a real use
case in the domain of human health and parasite biology. On the one
hand, we integrated the experimental data generated as part of an on-
going research on Chagas disease with the knowledge extracted from the
PubMed articles, using Semantic Web technologies. On the other hand,
we developed iExplore, a web tool with a graphical interface for inter-
active knowledge exploration, that allows non-technical users to explore
the integrated knowledge base using a relationship-focused approach. We
illustrate the effectiveness of our approach by describing the knowledge-
driven process of using iExplore to generate a new hypothesis for the
treatment of Chagas disease.
1 Introduction
Translating the knowledge from basic research into practice has been an impor-
tant trend in biomedical research in recent years. Translational research aims to
improve health by utilizing a wide range of biomedical resources, using knowl-
edge from experimental data at the point of care and guiding basic research with
problems encountered in patients. A large amount of biomedical knowledge is
available in the biomedical literature, e.g., in PubMed1 articles, and in structured
1
http://www.ncbi.nlm.nih.gov/pubmed
knowledge sources, including the Unified Medical Language System (UMLS) [2]
and the Entrez Gene database. Recent research [4, 6] has shown the potential
of using text mining on PubMed articles to advance the biomedical research by
exploiting the associations of genes and diseases from the text to prioritize the
candidate genes. However, we believe that these approaches do not exploit their
potential fully, because their context is limited by design to a specific subset of
the biomedical literature.
We argue that a broad knowledge base is key to enabling the generation of new
hypotheses. Using a large subset of the scientific literature will benefit biomed-
ical scientists performing basic or clinical research, and facilitate the integra-
tion of their data with the biomedical literature. We anchored our knowledge
base in comprehensive resources, such as the UMLS [2] and Entrez Gene. The
UMLS Metathesaurus enables the interoperability among data sources by pro-
viding reference identifiers for biomedical entities. Moreover, the UMLS Semantic
Network provides a consistent categorization of the UMLS Metathesaurus con-
cepts, together with a set of relations among the categories. The Entrez Gene
database contains the list of genes for various model organisms, as well as their
annotations. The combination of UMLS and Entrez Gene in the schema of the
Biomedical Knowledge Repository (BKR) provides the flexibility and scalabil-
ity for integrating additional data sources. We illustrate the integration of the
BKR with the experimental data obtained from the research in Chagas disease
in section 2.
We believe that the knowledge exploration process should be supported by a
tool that the scientists find easy to use and effective to help them gain new
insights from the integrated knowledge bases of text and experimental data.
The visualization should display sufficient contextual information to the users,
and the navigation should be driven by the intuition and background knowledge
of the scientists. We developed iExplore, a web tool that displays the graph
from integrated knowledge bases in an interactive manner using a relationship-
centric approach. The tool complements to the function of existing tools, e.g.,
RelFinder2 . While iExplore supports the exploration process, it is not supposed
to replace the role of the scientists in this process. We explain the knowledge
exploration process enabled by this tool in section 3.
2 Integration of knowledge sources
2.1 The Biomedical Knowledge Repository
The Biomedical Knowledge Repository (BKR) aims to integrate knowledge from
a variety of sources ranging from the scientific literature to various structured
knowledge bases [3]. The BKR contains relations extracted from PubMed docu-
ments by SemRep[1] and normalizes biological entities to concepts in the UMLS
and Entrez Genes. It includes approximately 20 million semantic predications
2
http://www.visualdataweb.org/relfinder.php
extracted from 6 million articles in PubMed published from 1999 to 2009. These
semantic predications are transformed into RDF format together with the prove-
nance information about the article where the predication is extracted.
Table 1. RDF triples represent “CALR associated with Chagas Disease”
Triple Subject Predicate Object
1 META C0041234 rdfs:label “Chagas Disease”
2 EG 811 rdfs:label “CALR”
3 META C0041234 INST rdf:type META C0041234
4 PUBMED 19108895/EG 811 rdf:type EG 811
5 PUBMED 19108895/EG 811 associated with META C0041234-INST
6 PUBMED 19108895/EG 811 derives from PUBMED 19108895
A semantic predication extracted from the title or abstract of an article is rep-
resented as a set of RDF triples. For example, the title “Trypanosoma cruzi
calreticulin: a possible role in Chagas’ disease autoimmunity” of the article with
PubMed ID 19108895 contains one predication, “CALR associated with Cha-
gas Disease”. The set of triples in Table 1 is created to represent this semantic
predication in the BKR.
2.2 Experimental Data about Chagas Disease
The experimental data about Chagas Disease originate from DNA microarray
analysis, proteome analysis, gene knockout and strain creation protocols. DNA
microarray analysis was used to measure the relative transcript abundances for
all of the genes in the T. cruzi genome during the four main life cycle stages of
T. cruzi, namely amastigote, trypomastigote, epimastigote and metacylic try-
pomastigote. Whole genome shot-gun proteomic analysis was used to measure
the presence of proteins encoded by T. cruzi genes during the four life cycle
stages. The proteome and transcriptome data have been used to prioritize genes
for the gene knockout and strain creation protocols. To capture the detail of
these experimental protocols, we created two ontolologies: Parasite Experiment
and Parasite Lifecycle, that have been published in the BioPortal3 . We use these
ontologies as schema to convert the experimental data into RDF.
2.3 Integration
A unified view of experimental data with biomedical literature requires the
mappings between entities of two data sources. Genes are one of the poten-
tial common entities between the BKR and experimental data sources. How-
ever, the experimental data are based on the Trypanosoma cruzi (T. cruzi)
3
http://bioportal.bioontology.org/
genome, while predications extracted from the biomedical literature refer to
human genes. The study of the human orthologs of T. cruzi genes is helpful
because the gene function is usually conserved across species. To bridge the gap
between genes of two organisms, we use the orthologous mapping from T. cruzi
to Homo sapiens(human) from the KEGG Sequence Similarity database4 , and
create an RDF triple for each pair of orthologous genes. For example, orthol-
ogy between the human gene “CALR” (ID 811 in Entrez Gene) and the T.
cruzi gene Tc00.1047053509011.40 is represented by the following RDF triple:
“EG 811 is orthologous Tc00.1047053509011.40.” In practice, such orthology re-
lations connect entities across the two sources.
3 Knowledge exploration
Knowledge exploration is the process of establishing a new relationship between
two known concepts and generalizes the notion of literature-based discovery
to sources others than the biomedical literature [5]. The exploration process
includes two steps, navigation and interpretation.
Navigation requires the visualization of knowledge as a set of relations. Such
relations can be explored by the user based on domain knowledge. We devel-
oped iExplore5 with an intuitive graphical interface to enable navigation. The
tool creates an abstraction layer from the RDF integrated knowledge bases to
visualize the graph of concepts and named relationships. Because knowledge is
represented as a set of relations, users can expand and narrow the graph based
on their interest. Graph expansion is implemented through predefined SPARQL
queries hidden from users. In practice, biologists drive the exploration guided by
their background knowledge, selecting concepts to expand the graph or restrict-
ing the graph to specific relations. Interpretation allows biologists to utilize
their background knowledge to generate novel hypotheses from the chains of
concepts identified in the navigation phase.
Example The exploration starts by expanding the concept “Chagas Disease”
and inspecting its related concepts. Of particular interest to us are known treat-
ments for Chagas disease. We restrict the graph to the “TREATS” relation
using a filter. Among the treatment concepts, we focus on drugs (categorized by
semantic types, such as “Pharmacologic Substance”). We find the drug itracona-
zole, known for treating various parasitic diseases and which has side effects. We
pursue our exploration by expanding the graph with the relations of the concept
“Itraconazole”. Specifically, we want to explore the genes connected to “Itra-
conazole” via the “INHIBITS” relation because they possibly indicate biological
pathways involved in the treatment of Chagas disease. Since only human genes
are present in the graph extracted from the literature, we follow the orthology
relation in order to find the T. cruzi orthologs of these human genes, i.e., the
4
http://www.genome.jp/kegg/ssdb/
5
Due to limited space, the tool and illustrated examples in section 3 are presented in
the tool’s homepage http://knoesis.wright.edu/iExplore/index.html for review.
possible target of the drug in the parasite. In summary, this example establishes
a chain of named relationships from “Chagas Disease” to itraconazole and hu-
man genes to T. cruzi genes. We generate a hypothesis from this chain, that
the T. cruzi orthologs of human genes inhibited by itraconazole may also be
inhibited by itraconazole and thus would be candidates for further studies into
the mechanism(s) of action of the drug itraconazole on T. cruzi.
4 Conclusion
We presented the semantic approach that underpins the integration and com-
plementary use of the two independently generated, heterogeneous knowledge
sources. These integrated knowledge bases together with iExplore allow biolo-
gists to use their knowledge to drive the exploration and generate new biomedical
hypotheses. The validation of such hypotheses is part of our future work. We also
plan to improve iExplore by learning the way biologists use their background
knowledge to generate hypothesis, and then automate the interpretation step
by making recommendations for hypothesis generation. Of note, our approach
can easily be generalized to other diseases and, more generally, to other data
sources integrated and explored together with the BKR following the approach
we demonstrated with Chagas disease. The two-step exploration process can also
be applied to make use of the broader integration of these independently gener-
ated knowledge sources.
Acknowledgements This research was supported by an appointment to the Re-
search Participation Program at National Library of Medicine, and the NIH R01
Grant number 1R01HL087795-01A1. We also acknowledge Dr. Thomas Rind-
flesch, Dr. Cartic Ramakrishnan, Dr. Priti Parikh, Jonathan Mortensen, Joshua
Dotson and Sarasi Lalithsena for help.
References
1. C. Ahlers, M. Fiszman, D. Demner-Fushman, F. Lang, and T. Rindflesch. Extracting
semantic predications from medline citations for pharmacogenomics. In Pacific
Symposium on Biocomputing, 2006.
2. O. Bodenreider. The unified medical language system (umls): integrating biomedical
terminology. Nucleic acids research, 32, 2004.
3. O. Bodenreider and T. Rindflesch. Advanced library services: Developing a biomed-
ical knowledge repository to support advanced information management applica-
tions. National Library of Medicine, 2006.
4. A. Faro, D. Giordano, and C. Spampinato. Combining literature text mining with
microarray data: advances for system biology modeling. Briefings in Bioinformatics,
2011.
5. D. Swanson. Fish oil, raynaud’s syndrome, and undiscovered public knowledge.
Perspectives in biology and medicine, 30(1):7, 1986.
6. N. Tiffin, J. Kelso, A. Powell, H. Pan, V. Bajic, and W. Hide. Integration of text-and
data-mining using ontologies successfully selects disease gene candidates. Nucleic
acids research, 33(5):1544, 2005.