As the amount of data being generated in biology has increased, a major challenge has been how to store and represent this data in a way that makes it easily accessible to researchers from diverse domains. Understanding the relationship between genotype and phenotype is a major focus of biological research. Various approaches to providing the link between genes and their functions have been undertaken - most require significant and dedicated manual curation. Advances in web technologies make possible an alternative route for the construction of such knowledge bases - largescale community collaboration. We describe here a system, the BOWiki, for the collaborative annotation of gene information. We argue that a semantic wiki provides the functionality required for this project since this can capitalize on the existing representations in biological ontologies. We describe our implementation and show how formal ontologies could be used to increase the usability of the software through type-checking and automatic reasoning.
There has been recent interest in the development of a
public knowledge base for information about genes and gene
functions [
To date the majority of gene annotation to biomedical ontologies has been performed by expert curators who use the public literature to determine the appropriate ontological terms for the annotation of genes and gene products. This method provides a highly accurate annotation, but is manually intensive and time-consuming. To overcome this bottleneck, automatic electronic annotation was also implemented, but this is generally less reliable than manual curation. The ability to tap the expertise of the large biological community potentially provides a stable, long-term strategy for maintaining and extending the information captured about gene functions. To this end we have developed the BOWiki, which provides a vehicle for the collaborative annotation of genes with concepts in the biomedical ontologies, as well as a suitable framework for the detailed description of functions as specified by OF.
The aim of the BOWiki is to capitalize on community
knowledge to build an accurate and easy-to-use knowledge
base about genes and gene functions [
We discuss here in some detail how information capture, storage, searching and quality assurance are managed inside the BOWiki.
We evaluated the suitability of various technologies to
serve as a framework for curating biomedical knowledge bases.
Due to our need for facilitating intuitive collaboration
editing, we chose a wiki as the base system for BOWiki [
The ability to edit biological functions in the BOWiki is realized through an implementation of the framework provided by OF. The OF treats functions as goal -oriented entities, and provides a formal way to represent functions in a function structure scheme. The scheme consists of a set of labels1, a set of requirements2, a goal 3 and a functional item4, which together form the function determinants. Via the function edit form (see figure 1), the function determinants are presented for editing. When the information concerning the function determinants is stored, concepts with function-, requirement- and goal-name are created or modified, together with the appropriate semantic links and structure. The complete content of the function determinants, as well as a link to the function edit form for further editing, is presented inside an information box on the function page. 2.3
The OF framework requires the use of n-ary relations. In
order to use OF and to allow for intuitive knowledge
acquisition, the BOWiki extends the Semantic MediaWiki to allow
n-ary relations between wiki pages. In order to enable the
use of n-ary relations, the BOWiki structures arguments of
relations into roles, or named argument slots. This allows n
arguments to be used in an arbitrary order. We believe that
this provides an intuitive way to describe complex relations
between entities in the BOWiki. This feature is not present
in most semantic wikis. For the semantics of n-ary relations
in the BOWiki, we use the proposal from the W3C Semantic
Web Working Group[
An example of a three place relation in BOWiki is the following function ascription (assumed to appear on the MAL21 wiki page describing the sugar transporting protein MAL21 ): [[realizes::function=sugar transporter activity; realization=sugar transport; context=human body]]
This means that the protein MAL21 realizes the function ”sugar transporter activity” (taken from the GO) by means of a process ”sugar transport” (taken from the GO) in the context of a human body. 2.4
In order to store the semantic content of the BOWiki, we
extended the Semantic MediaWiki’s database model. The
database model must accommodate n-ary relations, roles
and types from BOWiki’s type system (see section 4.1).
Therefore, a database layout that differs from the
triplerepresentation used in the Semantic MediaWiki [
The database model, as shown in figure 2, is subdivided into two parts, one focusing on classes (relations, roles and types), the other on instances (instances of relations, of roles and objects). Relations are instantiated each time they appear with a different set of arguments. These instances acquire an identifier from the relation instance table. The arguments table is comprised of the identifiers of the roles and objects used in a relation-instance. It is, therefore, a table of instances of roles. Using the relation-instance and argument table, the database layout allows storage of n-ary relations, and most operations can be completed by integer queries, which ensures rapid content retrieval from a large knowledge base.
For the incorporation of automated consistency checks, a table with available types will be added to the current database layout, and the types will be linked to the roles that may be used in a relation. This allows for the classification of objects occurring as arguments to relations in the wiki. 3. 3.1
Building a structured knowledge base with semantic annotations allows for improving regular search techniques through content based queries. In particular, in the context of gene annotations, it is important to be able to find correlations between genes, i.e., to see which gene has a specific relation to another, or which gene plays a certain role in a specific relation. In order to allow for content-based searches within the BOWiki, the Semantic Search Triple from the Semantic MediaWiki v0.4 has been expanded into a Search Tuple (see figure 3). This facilitates searching for an arbitrary number of terms of the following types: subject, relation, object, role and attributes. The search function employs numerical identifiers in order to increase its efficiency. This is in particularly significant when querying large databases, such as the GO, which contains more than 50,000 relations between concepts.
Inline Queries, as introduced by the Semantic MediaWiki, are adapted to the more complex representation schema of semantic information within BOWiki. Inline queries allow for the performance of semantic searches within wiki pages, representing the results as part of the rendered wiki page. 3.2
For referencing, browsing and as a starting point for local
articles, the BOWiki allows access to several ontologies like
the Gene Ontology [
Collaborative curation of a knowledge base may quickly
lead to semantic inconsistencies. A simple example of such
an inconsistency is the use of a part of relation between a
process and an object, which is often prohibited[
A core ontology provides definitions for the most basic
concepts within a domain. We developed GFO-Bio [
GFO-Bio incorporates higher order categories and employs an explicit instantiation relation, which are particularly relevant for its efficient use in a collaborative knowledge acquisition environment, where both concepts and individuals are described formally. Higher order categories allow biological concepts to be treated as instances instead of classes in OWL. We believe that this will also result in a more rapid and efficient classification.
All BOWiki pages, whether concepts or individuals, are instances of a class in GFO-Bio; we use instantiation instead of subsumption to increase the efficiency in reasoning. In some sense, reasoning is restricted to the core ontology because all wiki pages appear as OWL individuals for the reasoner. Consequently, subsumption links among concepts in the wiki are currently not exploited, which is appropriate for basic type checking and a tribute to the performance of current reasoners. Two types of inconsistencies may arise that can be avoided.
First, consider the BOWiki concept Hydrogen and the GFO-Bio concepts M olecule and Atom. In the BOWiki, Hydrogen may become an instance of M olecule. After instantiation, we want to prevent a user from making the Hydrogen also an instance of GFO-Bio’s Atom concept, because those concepts are disjoint in GFO-Bio. A molecule consist of atoms, but it is not an atom.
Second, concepts taken from GFO-Bio may serve as typerestrictions for roles in n-ary relations. Consider the realizes relation and its roles: the implicit subject role, the function, realization and context roles. These are restricted to the GFO-Bio concepts: Biological object, Biological function, Biological process and Entity. The use of a particular concept ”A” to fill the realization role of the realized relation causes this concept to be classified as a Biological process. If the same concept ”A” is used in the same relation, but in the subject role, it is then classified as both Biological Process and Biological Object, resulting in an inconsistency because in GFO-Bio, Biological Processes and Biological Objects are disjoint6.
Both types of inconsistencies are automatically detected within the BOWiki using the description logic reasoner Pellet6. The description logic reasoner Pellet7 is capable of performing these operations, and will be used for automated consistency checking in the next version of the BOWiki. 4.2
To further improve the quality of the articles, we are considering the inclusion of a content evaluation system, which may take the form of a direct evaluation of an article, or a cumulative score for an author’s body of work. Further research is needed in order to represent these ratings in the semantics of BOWiki’s content.
The BOWiki is still under development. The main task for the future will therefore be to enhance the user interface in order to make the BOWiki more user-friendly and intuitive.
We are considering integrating the BOWiki with a grid platform such as the MediGRID8, a portal for applications and web-services for biomedical research and the life sciences. Using web-services, the BOWiki may access biomedical ontologies directly rather than relying on daily updates. Web-services also provide the ability for users to perform queries of the BOWiki content within an integrated webservice platform.
For more restricted domains, a more expressive syntax may be required. For example, logical conjunction, negation or existential and universal quantification may be allowed. This may make it possible to extend the type system itself, instead of classifying every article as an instance of a concept in the type system. This will facilitate the collaborative construction of expressive ontologies. However, the main limitation will be the efficiency of the available reasoners, as well as the integration of a syntax for these constructs in the BOWiki.
To provide a graphical representation of the structured
content, we are considering conceptual graphs [
The BOWiki can be used to collaboratively create a
knowledge base about biological domains. It allows for the
specification and description of functions according to the
framework laid out by the Ontology of Functions [
9The interested reader can find a running system using the current stable release of BOWiki at http://onto.eva.mpg. de/bowiki