The notion of design patterns, after architect Christopher Alexander, provides a powerful way to capture and describe reusable design knowledge for complex domains. In this position paper we present the idea of design patterns for molecular biology experiments, and discuss how they may be utilized to support experimental design reuse, reproducibility, and a platform for linking experiments. Design patterns provide an alternate terminology and interpretive framework that can capture expert experience and intent that is critical yet missing from current representations of lab methods that utilize web ontologies and computational workflows. We outline an approach to making design patterns a first class entity in support of linked experiments on the web and provide a glimpse of potential applications of laboratory design pattern knowledge.
While there has been much focus on the description and linkage of scientific datasets
using web ontology languages, there has been less attention on how to describe and
then link the surrounding laboratory methods that are an important step in generating
such data. The use of biomedical ontologies1 to annotate laboratory methods
descriptions is a much needed and necessary first step to integrating laboratory experiments,
however, current ontologies alone cannot always provide sufficient knowledge to
support all the human reasoning and situated understanding one may need to act with
such knowledge [
Design patterns were first introduced in the domain of architecture by Christopher
Alexander [
Fig1. Design patterns to capture laboratory knowledge.
Patterns not only give a set of concepts and vocabulary for a domain, but do so
in a way that tells us what to do.2 The problem/solution orientation of patterns gives
us metaphorical dials to the domain, and tells us how to control and operate
effectively with them. Unlike ontologies which aim to be capable of expressing any valid
domain knowledge, patterns act more like recipes (yet more abstract and general than
typical workflow representations) and give us a map3 of the model space that tells us
what parts we can vary, and what should remain invariant to achieve the desired
outcome [
Other patterns exist in the context of the semantic web e.g. workflow4 or ontology design patterns5, which aim to provide usually domain independent constructs for normalizing and specifying knowledge modeling problems. In contrast, our notion of laboratory patterns as knowledge acquisition abstracts over laboratory procedures directly (cf. the modeling of them) to provide reusable design solutions for scientific experiments anchored in domain context - they supply us with domain concepts and relationships across diverse experiments gathered around a specified design intent. 2 For examples of lab patterns and use cases see http://goo.gl/D5RZsQ 3 A map analogy of patterns at http://hillside.net/plop/2010/papers/kohls.pdf 4 http://www.workflowpatterns.com/ 5 http://ontologydesignpatterns.org/wiki/Main_Page
While we accentuate differences in representational approaches here, in reality we recognize the boundaries between ontologies, workflows, and design patterns are fuzzy as each tries to incorporate aspects of the other. For the purpose of discussion we make some general distinctions between the traditional forms of the three in Table 1 below.
The traditional form of design patterns are structured documents written in natural language. Thus, in order to transform them into a resource for linked science, a mechanism for publishing patterns and their vocabulary as defined web addressable entities is desired.
We view design patterns as data and ask how we may publish pattern
knowledge following linked data practices. To begin we have developed a method for
capturing pattern knowledge utilizing social methods adapted from other domains
where design patterns are valid entities. The structured documents that result from
“pattern mining” are collaboratively transferred to a semantic wiki based on the
OntoWiki Application framework [
The challenges to this approach consist of specifying and refining the semantic formalization of pattern level concepts and their relations using RDFS, OWL and appropriate logics, and subsequently tailoring the OntoWiki Application framework. This work is non-trivial as patterns have complex, interrelated internal and external structures, and remains the current focus of our efforts.
A vision for the application of laboratory pattern knowledge
The annotation of lab procedural descriptions with vocabulary and context supplied
by patterns is an obvious use of patterns as metadata. Currently, this coupling must be
created manually due to the implicit nature of many pattern concepts, but the markup
of existing documents or the authoring of future ones can be facilitated by adapting
existing annotation tools such as Rightfield [
Patterns name invariant forces that exist in recurring lab scenarios and provide
a valuable step towards the specification of minimal information reporting guidelines
for diverse laboratory processes. Indeed, the need for “high-level abstractions of the
components of experimental workflows” has been noted [
Our vision is the creation of a laboratory pattern catalogue and web resource, providing scientists assistance in understanding, reusing, and adapting the diversity of published laboratory methods to their own needs. Principal in our approach is the publication of pattern content and vocabulary as linked data, such that it may be available for use anywhere on the semantic web.
We believe the problem/solution orientation of design patterns fits well with the cognitive processes of laboratory scientists when engaged with methods knowledge. In combination with workflows and ontologies, the pragmatic aspects of pattern knowledge can help provide a type of balancing – filling a representational gap in our methods descriptions somewhere between axiomized ontologies and workflows that can improve the epistemological adequacy of our scientific record. 5