BioMixer: Visualizing Mappings of Biomedical Ontologies Elena Voyloshnikova*, Bo Fu, Lars Grammel, Margaret-Anne Storey Department of Computer Science, University of Victoria, Canada ABSTRACT more than two ontologies, or a way to visualize which The majority of current ontology mapping visualization tools ontologies have mapping in a large set of ontologies. In this are limited to showing mappings between a pair of ontologies at a paper, we present how the BioMixer tool (Fu et al., 2012), a time. However, it is often the case that concepts from one ontology are mapped to concepts in several other ontologies. Understanding tool for visualizing ontologies, has various techniques for how multiple ontologies relate to one other, as well as understand- visualizing mappings at both the term and at the ontology ing the quality of mappings created across ontologies, can be sup- level across multiple ontologies. ported through visualizations that show mappings across more than two ontologies. In this paper, we present how BioMixer, a tool for 2 BIOMIXER MAPPING VISUALIZATIONS visualizing biomedical ontologies, provides a number of custom- izable views to support the understanding, analysis and navigation Through discussions with ontology users and ontology of mappings across multiple ontologies. curators, we became aware that visualization of mappings at the term level, as well as aggregated mappings at the term 1 INTRODUCTION and ontology level, would be desirable. Thus, BioMixer With a rapid growth of the semantic web, there is an contains three visualizations that show mappings between increasing need to visualize ontologies as well as to multiple ontologies. These three views differ in their level visualize how ontologies that are somehow related may of detail and in their scalability. The mapping overview have concepts mapped to each other (Falconer & Storey, aggregates mappings between a large amount of terms at the 2007). For example, in the NCBO1 BioPortal repository, ontology level, the mapping matrix shows mappings there are many mappings stored between terms in the between many terms at the term level ordered by ontology Protein Modification ontology and the PRotein Ontology or term label, and the detailed mapping graph shows the (PRO), but there are also 423 terms in the Protein mappings between a few terms within the context of other Modification ontology that are mapped to the Chemical term relationships. entities of biological interest ontology. A potential user of The mapping overview visualization (Fig. 1) provides a any one of these ontologies, may wish to gain an summary of mappings across multiple ontologies. When understanding of how all three ontologies are related by there are a large number of ontology term mappings, it is mappings, or an ontology curator may wish to explore how difficult to show that much information in detail. The user the three ontologies are mapped and whether such mappings can use this overview visualization to decide which make sense or are incomplete. ontologies and terms are relevant for viewing their Although visualizing mappings among multiple mappings. For example, the content of the mapping ontologies can provide valuable information, most existing overview can be based on a keyword search for terms across visualization tools that show ontology mappings, confine multiple ontologies. With this view, the user can quickly see the user to view exclusively two ontologies at a time from a which ontologies have many or lack any mappings. The single perspective. Some of the most common approaches next two views allow the user to drill in to explore for mapping visualization include (1) visualizing two mappings in detail. ontologies side by side and showing similarities visually in The mapping matrix visualization (Fig. 2) facilitates terms of matching position, colour, shape, or pattern to the understanding of mapping patterns at the term level. show the alignment, as in AlVIz (Lanzenberger & Sampson, Terms can be ordered by either term label or by ontology 2006) and Optima (Kolli & Doshi, 2008), and (2) showing name. Users can easily see clusters of mappings for indented trees for two ontologies where mappings are similarly-named terms or for ontology, and thus identify represented by links connecting matching terms between the potentially missing mappings. The matrix visualization also two ontologies, as in CogZ (Falconer & Storey, 2007) and supports understanding how a subset of concepts from one COMA++ (Aumueller et al., 2005). ontology is mapped into a set of other ontologies. What is lacking, however, is a visualization tool that can The detailed mapping graph (Fig. 3) supports users in show mappings or clusters of mappings across terms in analyzing and understanding mappings in the context of other term relationships. The user can search for a term of interest using the BioMixer search feature. The results can * To whom correspondence should be addressed: elenavoy@uvic.ca be showed in the detailed graph view, with mappings and 1 http://www.bioontology.org/ 1 Voyloshnikova et al. other relationships expanded. For example, the user can view the parent-child hierarchies and mappings for terms from multiple ontologies. Such a view should help a user in identifying missing mappings along the hierarchies, as well as gaining a broader idea about the meaning of similar or unrelated but superficially similar terms. Providing three mapping visualizations with different levels of detail allows BioMixer to address a variety of use cases, which is hard to achieve with a single visualization. The three visualizations can be linked together to implement the visual information seeking mantra (“Overview first, zoom and filter, then details-on-demand” (Shneiderman, 1996)) in the domain of ontology mapping visualization. Fig. 3. Detailed Mapping Graph. Concepts are shown as nodes, which are colored by ontology. Both mappings between ontologies (dashed gray edges) and relationships within one ontology (solid blue edges) are displayed. Users can expand mappings and other relationships on each node. Different layouts (e.g. force-directed, circle, tree) can be applied. 3 CONCLUSION The BioMixer tool was designed to support the visualization of mappings across multiple ontologies. It uniquely visualizes mappings in a variety of ways that has not been previously supported by other ontology mapping visualization tools. Early feedback indicates these views will be useful for exploring, analyzing and editing mappings Fig. 1. Mapping Overview Visualization. The ontologies are shown as across multiple ontologies. Future work involves evaluating circles. The radius represents the number of concepts in the ontology. The these views within the NCBO BioPortal website, and number of mappings between two ontologies corresponds to the width of developing future views based on user feedback. the edge between them. REFERENCES Aumueller, D., Do, H.-H., Massmann, S., and Rahm, E. (2005). Schema and ontology matching with COMA++. In: Proceedings of SIGMOD, 906–908. Falconer, S. and Storey, M.-A. (2007). A cognitive support framework for ontology mapping. In: Proceedings of ISWC/ASWC. Fu, B., Grammel, L., and Storey, M.-A. (2012). A Web-based Collabora- tive Ontology Visualization Tool. To appear in: Proceedings of ICBO 2012. Kolli, R. and Doshi, P. (2008). Optima: Tool for ontology alignment with application to semantic reconciliation of sensor metadata for publica- tion in sensormap. In: ICSC, 484–485. Lanzenberger, M. and Sampson, J. (2006). AlViz - a tool for visual ontol- ogy alignment. In: IV 2006: Proceedings of the conference on Informa- tion Visualization, Washington, DC, USA: IEEE Computer Society Press, 430-440. Shneiderman, B. (1996). The Eyes Have It: A Task by Data Type Taxono- Fig. 2. Mapping Matrix Visualization. Concepts from multiple ontologies my for Information Visualizations. In: Proceedings of the IEEE Sym- are shown as rows and columns. Mappings are shown as colored cells. The posium on Visual Languages,Washington, DC, USA: IEEE Computer user can hover over mappings to highlight row and column. The concepts Society Press, 336-343. can be sorted by name or by ontology. 2