Lessons Learned - User-Level Workflow Design: A Bioinformatics Perspective - page 151

domain model, which is not irrefutable after the initial domain modeling pro-
cess: The workflow designer is free to adapt the domain model according to
his concrete needs and bring in his specific expert knowledge at any time.
This is in fact specifically supported by the PROPHETS plugin: the user can
change the service interface descriptions, edit the service and data type tax-
onomies, and remove and add constraints via the graphical user interface. As
detailed in Section 8.1.4, this extremely flexible domain modeling approach
is so far unique among the frameworks for automatic workflow composition
in the bioinformatics domain.
Precise Domain Vocabulary
For the semantic service interface descriptions, a domain vocabulary is re-
quired that provides a precise terminology to describe and clearly distinguish
the services and data types in the domain. Especially in scientific applications,
the data that is processed by the services in the workflows is central, and
accordingly it is important that the domain vocabulary comprises domain-
specific terminology for referring to data. Since in the loose programming
framework the data types in the semantic domain model are symbolic and
completely decoupled from the technical data types of the underlying SIB-
s/services, the domain modeler can in fact deliberately define the data types
that he considers adequate for the synthesis application.
Clearly, existing ontologies or taxonomies should be reused (if available).
The currently most popular example of a readily available (technical) ontol-
ogy in the bioinformatics domain is EDAM (cf. Section 3.3.1), which provides
a comprehensive vocabulary for the annotation of all kinds of bioinformatics
entities. However, if the existing vocabularies are not precise enough, they
should simply be extended or replaced by more concise terminology. As de-
tailed in the following, the definition of the domain vocabulary varied consid-
erably between the four application scenarios of this topic.
Owing to community efforts, the EMBOSS tool suite, which forms the ba-
sis for the first loose programming application scenario presented in Chapter
3, has already been completely annotated in terms of the EDAM ontology.
Thus, the service and type taxonomies of the respective PROPHETS do-
main model could simply be derived from the corresponding branches of the
EDAM ontology, and the services from the EMBOSS suite could be char-
acterized and sorted into the taxonomies according to the available EDAM
annotations. As demonstrated in detail in Section 3.3, this domain model pro-
vides a proper basis for effective constraint-driven synthesis of phylogenetic
analysis workflows.
For GeneFisher-P (the workflow scenario discussed in Chapter 4), which
is also concerned with several sequence analysis steps, it was again possi-
ble to use the EDAM ontology as basis for the taxonomies in the domain
model. Semantic categories for the involved computational services (such as
PCR primer design for the GeneFisher service) as well as for the associated