August 8, 2007 | When working with data, it is just as important to visualize it properly as it is to process it, especially in the context of aligned (columnar) collections of records, such as HTS assay, microarray, and clinical data. Today’s researchers have an arsenal of visualization tools at their disposal, ranging from spreadsheets, statistical graphs, dynamic web pages, network views, heat maps, sector maps, and many more. In most cases these vu-tools work best with homogenous, flat data. But in the world of the Semantic Web, the data is often heterogeneous and complex in structure, and so the current ensemble of tools will only work for select slices of RDF data. The real trick is how to take advantage of, and visualize highly interconnected and typed data in a way that allows users to easily choose the perspective that makes most sense.
For many people there is the assumption that if one gains semantics, one has to lose visual understanding, which is why most of us like web pages. It’s as if by making data machine-readable, we worry we may lose human comprehensibility. Yet there are many ways to see data and work with it, so in principle, if semantics can help describe the data, then viewing options should only get more flexible and compelling. What is really needed is a standard way for browsers to easily know how to handle certain kinds and sets of semantic data that come in the form of RDF — the same way style-sheets are used for HTML and XML files.
One tool that offers an intuitive and natural way to view semantic data is EXHIBIT (http://simile.mit.edu/exhibit/), developed as part of MIT’s SIMILE project (http://simile.mit.edu/). EXHIBIT has two noteworthy features: it does not require any back-end database (relational or RDF triple-store) that an author needs to intricately connect to; and it allows anyone to render RDF data via a standard browser by using simple HTML data templates and a set of pre-defined viewing modalities: tile views, tables, timelines, scatter plots, and bar graphs. What’s so cool about EXHIBIT is that once you’ve pointed it to your RDF file or data feed, it dynamically creates interactive ‘facets’ that you can use on the web page to select subsets of the data. Facets are selectable lists of types that a data-record may contain, such as “SMOKER” vs. “NON-SMOKER” for SMKCLASS or “MILD” vs “SEVERE” for “ADVERSE EVENT”. Facets allow a fast and easy way to filter those records having the desired combinations of values types. Although EXHIBIT is based on JavaScript, it is contained so you don’t have to call or write any additional JavaScript — just include an attribute template tag for the desired data properties (columns) contained in our RDF data records:
<div ex:role=”exhibit-view”
ex:viewClass=”Exhibit.TabularView”
ex:label=”Subject Demography”
ex:columns=”.label, .AGE, .sex, .wt, .ht, .smkclass, .AE “
ex:columnLabels=”Subject, Age, Sex, Weight, Height, Smoker, Adverse Events”
ex:sortColumn=”1”
ex:possibleOrders=”.AGE”
ex:sortAscending=”true” >
</div>
From this set of cues, EXHIBIT is able to generate a fully styled HTML representation of all your data as a table, nicely formatted by rows and columns, where the RDF predicates you want as columns are specified under ex:columns. To create facets, all you specify is:
<td width=”20%” style=”font-size:10px”>
<div ex:role=”facet” ex:expression=”.sex”></div>
<div ex:role=”facet” ex:expression=”.smkclass”></div>
<div ex:role=”facet” ex:expression=”.AE”></div>
<div ex:role=”facet” ex:expression=”.treatment”></div>
</td>
A useful companion technology is SIMILE’s Babel translator (http://simile.mit.edu/babel/), an on-line converter that can take your data, even spreadsheet files, and convert them into authentic RDF. I have used it to take several heterogeneous files of clinical trials data (synthetic), and merge them all into one RDF document that now can be easily viewed using EXHIBIT. Using this approach I can combine standard table documents from clinical trials and view them in a variety of pivotable ways (http://eneumann.org/exhibit/clinicaldemo/). Babel can also convert files, like json (JavaScript Object Notation), from other formats as well, but I should mention it seems to work best with a Firefox browser. One can also call Babel via a web-request, circumventing the need to manually go through the web page.
Another system for viewing RDF data is TABULATOR (www.w3.org/2005/ajar/tab), an ongoing project by Tim Berners-Lee and his team to explore linked semantic data. This tool can be added to Firefox browsers as a plug-in, and allows users to explorer complex sets of RDF data in a click-able navigational way that downloads and renders all the linked relations as an expanding tree. Since Semantic Web objects form a graph, the tree view often times descends to a point that links back to something at the top-many find this a bit confusing. TABULATOR is able to create GoogleMaps if location data is part of the data set, and supports SPARQL queries. It is a tool designed for the technically savvy Semantic Web surfer.
Lastly, there is an effort to develop visualization standards for the Semantic Web called Fresnel — pronounced fre-nel’ (www.w3.org/2005/04/fresnel-info/). Fresnel defines a vocabulary set on how to select, render and view RDF data so that it will work in different browser environments. To date, several browsing systems implement Fresnel: Cardovan (IBM), IsaViz (W3C/INRIA), Longwell / Piggy Bank (SIMILE/W3C/MIT), Arago (DERI), and Horus (Freie Universität Berlin). Semantic Web visualization is still a topic of research, but as more semantic data comes on-line, the use and demand for such tools will grow proportionally. For now, there are some practical, easy to use prototypes that will enable researchers to better understand the advantages of Semantic Web information.
Eric K. Neumann is senior director product strategy at Teranode. E-mail: eneumann@teranode.com.
Subscribe to Bio-IT World magazine.