While there are many technical approaches to building wiki software, the one component that is crucial to success is a large, motivated community of users and editors. Generating such a community from scratch is a difficult, complex process that is not well understood and often does not succeed. Because of this challenge, several groups decided to focus their efforts in the context of a community that is already large and active—Wikipedia (3, 4). Rather than building a novel wiki and then working to attract users, these initiatives started with a very large user base and worked to help these users expand and improve the content of the articles relevant to their initiatives. This choice has largely resulted in successes in terms of high quality, volunteer-based content production (5, 6); however, operating in the context of Wikipedia entails a lack of control of the implementation and thus limits what can technically be achieved.

In addition to aiding navigation for users, hyperlinks may implicitly indicate meaningful relationships between concepts. Where hypertext documents are principally about a particular concept, as is the case in Wikipedia, links between documents provide a loose indication that the concepts represented by those documents are related. In the example above, we may infer the possibility that the 5-HT1A receptor protein plays a role in the process of vasodilation based simply on the presence of the hyperlink connecting the two concepts (7). However, such inferences are imprecise. In the best case, the nature of the relationship between the concepts is undefined (e.g. there is no indication how the 5-HT1A receptor is related to vasodilation) and in the worst case it is non-existent (e.g. the link appeared in the context of a sentence that was not about the 5-HT1A receptor).

Semantic Links make it possible for editors to explicitly specify the meaning of the links in a hypertext document. This construct is the foundation of the Semantic Media Wiki (SMW) system, an extension of the basic MediaWiki framework intended originally to enable the creation of a ‘Semantic Wikipedia’ (8, 9). Within an SMW system, an editor can specify the nature of the relationship between concepts represented on the wiki and the system can then use that additional information to improve the organization of the site. For example, an editor can state that San Diego is a city in California and the system can then use this information to respond to queries like ‘list all cities in California’. Without the addition of the ‘city in’ relationship, such a query would be extremely difficult if not impossible to answer.

The MediaWiki system translates the WikiText used by editors into standard HTML understood by Web browsers. During this translation process, WikiLinks like [[Protein kinase A]] become standard HTML hyperlinks like <a href=“/wiki/Protein_kinase_A”>Protein kinase A</a>. SWLs are translated into the following structured HTML (modeled after the ‘microformat’ pattern (microformats.org):

<span class=“swl”>

<span class=“Phosphorylated_by”>

<a href=“/wiki/Protein_kinase_A”>

Protein Kinase A</a>

</span>

</span>

We provide two Wikipedia userscripts, one that uses embedded SWLs to enhance the presentation of the articles where they occur and another to ease the process of inserting them when editing an article. The former generates a tab that, when clicked, renders an infobox displaying all of the SWLs discovered on the page (Figure 3). This table provides a rapid summary of the key attributes of the concept represented by the page (e.g. a gene) and allows the user to quickly navigate to linked articles and to the relationship types used in the SWLs. The other script adds a button to the editing toolbox that helps to produce a correctly formatted SWL (Figure 4).

During the transfer process, the system detects SWL templates in relevant Wikipedia articles and converts them to semantic links as they are represented in the Semantic Media Wiki framework. For example, the article on adenosine deaminase contains the following SWL:

{{SWL | target=hemolytic anemia | type=overexpression_results_in}}

When the article is transferred to Gene Wiki+, the system translates this to the Semantic MediaWiki equivalent:

[[overexpression_results_in::hemolytic anemia]]

This translation has two immediate consequences in the context of Gene Wiki+. First, the semantic link is visible through the ‘browse properties’ feature when viewing either the adenosine deaminase article or the article on hemolytic anemia (Figure 5). Second, the relationship can be used in queries such as ‘list all the genes whose overexpression results in hemolytic anemia’. These semantic features are enabled as default behaviors of the Semantic Media Wiki system and hence accessible through the Gene Wiki+ instance. In addition, the semantic relationship type for the SWL is brought over and encoded as a Semantic Media Wiki Property. This allows Gene Wiki+ users to view its textual definition and to navigate to related properties within external vocabularies.

As illustrated in Figure 5, the Gene Wiki+ system contains semantic links translated from SWLs in Wikipedia such as ‘mutation results in’ and ‘overexpression results in’, but it also contains a large number of ‘is associated with’ relationships. The system generates these loosely typed relationships whenever a standard (i.e. untyped) hyperlink is detected in the source article on Wikipedia. While it will take time for SWLs to be inserted into Wikipedia, these simple associations enable interesting queries immediately. For example, a simple hyperlink to the article on phosphorylation is translated to [[is_associated_with::phosphorylation]]. Genes that code for kinases, for instance, nearly all include this particular link; translating it to a semantic link allows us to search for all genes that are associated with phosphorylation. The semantic query can then be expanded to find genes that are involved with phosphorylation and also in cancer, and so forth. These queries will become more powerful as these generic ‘is associated with’ relationships are made more precise (‘catalyzes’ for kinases, ‘catalyzes removal’ for phosphatases, etc.).

Aside from its use in direct association queries, the ‘is_associated_with’ property also forms the root of the property hierarchy used in Gene Wiki+. This means that queries for X where X ‘is_associated_with’ Y will return results where X ‘some more specific relationship’ Y. Figure 6 illustrates how simple associations and more specific semantic relationships can be blended in a Gene Wiki+ query. The table returned by the query lists all genes related somehow to hemolytic anemia. In addition, it presents specific semantic relationships, such as ‘overexpression results in’ or ‘biomarker for’ that link each gene to hemolytic anemia and to other diseases.

The process of transferring content from Wikipedia to Gene Wiki+ runs continuously such that all changes to relevant articles on Wikipedia are brought over to the Gene Wiki+ in near real time. This makes it possible for users to edit Wikipedia articles with SWLs and then immediately see how those changes impact queries in Gene Wiki+. The articles brought over from Wikipedia are not editable on the Gene Wiki+; any changes should be made directly on the original Wikipedia article.

To facilitate integration with other Linked Data resources, all Gene Wiki+ articles are annotated with their equivalents in DBpedia (13). DBpedia is a Web-accessible RDF database constructed automatically by extracting data from the few consistently structured parts of Wikipedia pages such as categories and ‘infoboxes’. Infoboxes display facts as key-value pairs rendered in a table typically visible on the upper right of Wikipedia articles. For example, there is a taxonomy infobox that contains data about the scientific classification of organisms (e.g. kingdom, phylum, etc.). Where available, such data is extracted automatically and represented in the DBpedia system. The DBpedia system makes no attempt to extract data from the hypertext of Wikipedia articles and, as such, is entirely complementary to the data represented in embedded SWLs. Mapping Gene Wiki+ to DBpedia provides the opportunity to very easily integrate the information derived from SWLs embedded in article text with what structured data does exist on Wikipedia. In addition, DBPedia is a central point for ontology term mapping for the Semantic Web and, as a result, integration with it begins the process of integration with the many other knowledge bases that also map their concepts to it.

Aside from DBpedia integration, the Gene Wiki+ also produces equivalency links between relationship types in SWLs and properties in external ontologies. This is achieved by processing SWLs that appear on relationship type pages in Wikipedia and detecting when they contain the ‘equivalent’ relationship. For example, the relationship type page for ‘biomarker for’ contains the SWL:

{{SWL | type=equivalent

| target=http://ncicb.nci.nih.gov/xml/owl/EVS/ Thesaurus.owl#Gene_Is_Biomarker_Of

}}

Here, we implement a partial solution by decoupling the encoding of semantic relationships from the querying and utilization of those relationships. We provide a mechanism for the massive Wikipedia community to participate in the process of assembling not just a textual resource, but a far more powerful structured knowledge base. Mining and utilizing that knowledge base can then be performed using third-party tools without requiring changes to Wikipedia itself.

Of course, this approach is not without its challenges. Since anyone can create a SWL and there is no way to enforce the use of a particular set of relationship types, it is certain that there will be some inconsistency in how they are used. For example, it is possible that one editor might use the relationship type ‘phosphorylated by’ while another editor might insert ‘phosphate group added by’. Such inconsistencies disrupt semantic queries made over aggregated content such as those available through the Gene Wiki+. However, this is no different from other aspects of Wikipedia articles. Over time, the Wikipedia community has gradually moved toward consensus regarding the use of categories, when to insert normal WikiLinks, when to insert references and how to format them, how to format articles and even how to style the text in the articles. Since Wikipedia is a continuously changing, social artifact there will always be exceptions to the socially defined rules that have emerged to govern its content, but overall, the basic structures remain remarkably stable. If the Wikipedia community takes up SWLs, we expect the same kind of social consensus to emerge with respect to their use. The community will evolve rules defining which properties should apply to which kinds of entities and will police the articles for adherence to these rules in the same way they do now for other article attributes. The key question is whether the community will in fact buy in to the SWL idea.

Right now, there are only a handful of SWLs active on Wikipedia, all of which have been deposited by our team as demonstrations to seed the process. If the SWL concept succeeds, this number should rapidly increase into the tens of thousands, but for that to happen, the Wikipedia editor community must become involved as semantic link authors. In order to recruit this labor, the value of such work needs to be clearly apparent. Since Wikipedia itself does not process the semantic relationships, external applications need to be developed and promoted that make the value of these contributions clear, thereby providing editors with vital positive feedback. Applications like Gene Wiki+ and the infobox-generating userscript provide some first steps in this direction. These applications demonstrate that adding semantic relationships can substantially enhance users interactions with the knowledge in Wikipedia and can enable the production of novel applications relevant to biocuration activities.