Conclusion - User-Level Workflow Design: A Bioinformatics Perspective - page 192

• Constraints that describe the overall structure of a workflow, like requiring

that the services X , Y and Z have to be included in exactly this order.

The first kind of constraints can on the one hand already be provided by
the domain modeler or later by the workflow designer, capturing the specific
user's expert knowledge about the types and services in the domain. On
the other hand, such information could systematically be included in the
service documentations of the providers, which would prevent the redundant
identification and formalization of the respective information by different
users.
The second kind of constraints can of course be defined directly by the
workflow designer, who uses constraints to describe the overall structure of
the intended workflow. However, it would be even more interesting to pro-
vide ready-to-use (constraint) libraries of domain-specific workflow patterns,
that is, templates of frequently occurring bioinformatics service compositions.
This goes beyond the general, domain-independent workflow patterns iden-
tified and described in [326]. Possible sources for the systematic, large-scale
identification of such domain-specific patterns are, for instance:

• Event logs of major service providers (such as the European Bioinformat-

ics Institute, EBI) or service portals (such as BioMoby [341]), in which
process mining frameworks (like, e.g., ProM [337, 328, 51]) could find
several patterns,

• Workflow provenance data , that is, metadata that helps managing the

large amounts of data in the course of scientific experiments [267], which
in contrast to the mere event logs allows for a better discrimination of the
data dependencies [354], and

• The has input and has output relations that the EDAM ontology (cf. Sec-

tion 3.3.1) defines for the Operation terms in the ontology. Simply regard-
ing the Operation terms as services, the synthesis methods of Bio-jETI,
for instance, might easily find service sequences in the abstract terms of
the ontology.

In addition to being used as workflow constraints by domain modelers and
workflow designers, domain-specific workflow patterns could also be useful for
improving the performance of the search for adequate workflows, for instance
by their application for domain-specific search heuristics (see above).
In the same way as EDAM is a publicly available constituent of bioin-
formatics domain models, additional domain-specific knowledge as described
above should ideally also be made available via a common public repository.
Such a library of constraints that can be added and removed dynamically
during workflow development would enable users to work and experiment
with the domain in a very flexible manner. The details of such a platform
would have to be negotiated within the community, but existing Semantic
Web technology could certainly be employed for this purpose.