Conclusion - User-Level Workflow Design: A Bioinformatics Perspective - page 184

design as in Bio-jETI provides means for reasoning about workflow prop-
erties at the user level.

•

Constraint-driven workflow design extends the application of constraints
further, systematically using them for high-level descriptions of individual
components and entire workflows, which can then be translated automat-
ically into concrete workflows that conform to the constraints by design.
The PROPHETS plugin supports the synthesis of workflows based on
specifications in terms of temporal-logic constraints according to the loose
programming paradigm. In particular, it allows users to describe the in-
tended workflow at an abstract, less technical level by providing simple
means for sketching workflow “skeletons” and for expressing workflow con-
straints in terms of domain-specific vocabularies. It is applied for (semi-)
automatic workflow design within Bio-jETI.

Bio-jETI's workflow design approach has been applied to a number of
academic and real-life bioinformatics workflow scenarios. The application ex-
amples presented in this topic cover different thematic areas (phylogenetic
analyses, PCR primer design, metabolic flux analysis, and microarray data
analysis), different software components (commonly known tool collections as
well as special-purpose services), different service technologies (standard Web
Services as well as especially created jETI services), and also workflows of
different complexity. Their different characteristics allowed for the assessment
of the current implementation in particular and provided insights regarding
the capabilities and limitations of the approach in general.
Based on the experiences gained by working on the application scenar-
ios and the results of the evaluation, a set of loose programming pragmatics
has been formulated that provide general guidelines for adequate domain
modeling and synthesis application. With regard to domain modeling, they
emphasize the importance of adequate granularity of service and data types,
and of a precise domain-specific vocabulary and clear semantic service inter-
face descriptions in the domain models. With regard to the actual workflow
design, they advocate an incremental exploration of the solution space, until
the intended or an adequate set of solutions is obtained.

9.1.2 Remaining Challenges

There is of course room for improvement of the current implementation of
the framework, but for many issues it is already relatively clear how they
can be addressed, so that solving them is more or less a matter of doing
(cf. Section 9.2). However, for two open issues that are crucial for the future
success of the method, namely the performance of the synthesis algorithm
and the origination of “good” domain models, the answers are not as clear,
so that two major challenges remain.