Knowledge gained from the scienti c literature can complement newly obtained experimental data in helping researchers understand the pathological processes underlying diseases. However, unless the scienti c literature and experimental data are semantically integrated, it is generally di cult for scientists to exploit the two sources e ectively. 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 exploration 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 ongoing 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 interactive knowledge exploration, that allows non-technical users to explore the integrated knowledge base using a relationship-focused approach. We illustrate the e ectiveness of our approach by describing the knowledgedriven process of using iExplore to generate a new hypothesis for the treatment of Chagas disease.
Translating the knowledge from basic research into practice has been an
important trend in biomedical research in recent years. Translational research aims to
improve health by utilizing a wide range of biomedical resources, using
knowledge 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 Uni ed Medical Language System (UMLS) [
We argue that a broad knowledge base is key to enabling the generation of new
hypotheses. Using a large subset of the scienti c literature will bene t
biomedical scientists performing basic or clinical research, and facilitate the
integration of their data with the biomedical literature. We anchored our knowledge
base in comprehensive resources, such as the UMLS [
We believe that the knowledge exploration process should be supported by a tool that the scientists nd easy to use and e ective to help them gain new insights from the integrated knowledge bases of text and experimental data. The visualization should display su cient 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 relationshipcentric 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 2.1
The Biomedical Knowledge Repository (BKR) aims to integrate knowledge from
a variety of sources ranging from the scienti c literature to various structured
knowledge bases [
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 trypomastigote. 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
A uni ed view of experimental data with biomedical literature requires the mappings between entities of two data sources. Genes are one of the potential common entities between the BKR and experimental data sources. However, 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, orthology 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 relations connect entities across the two sources. 3
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 [
Navigation requires the visualization of knowledge as a set of relations. Such relations can be explored by the user based on domain knowledge. We developed 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 prede ned SPARQL queries hidden from users. In practice, biologists drive the exploration guided by their background knowledge, selecting concepts to expand the graph or restricting the graph to speci c relations. Interpretation allows biologists to utilize their background knowledge to generate novel hypotheses from the chains of concepts identi ed 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 treatments for Chagas disease. We restrict the graph to the \TREATS" relation using a lter. Among the treatment concepts, we focus on drugs (categorized by semantic types, such as \Pharmacologic Substance"). We nd the drug itraconazole, known for treating various parasitic diseases and which has side e ects. We pursue our exploration by expanding the graph with the relations of the concept \Itraconazole". Speci cally, we want to explore the genes connected to \Itraconazole" 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 nd 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 human 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
We presented the semantic approach that underpins the integration and complementary use of the two independently generated, heterogeneous knowledge sources. These integrated knowledge bases together with iExplore allow biologists 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 generated knowledge sources.
Acknowledgements This research was supported by an appointment to the Research Participation Program at National Library of Medicine, and the NIH R01 Grant number 1R01HL087795-01A1. We also acknowledge Dr. Thomas Rindesch, Dr. Cartic Ramakrishnan, Dr. Priti Parikh, Jonathan Mortensen, Joshua Dotson and Sarasi Lalithsena for help.