Lessons Learned - User-Level Workflow Design: A Bioinformatics Perspective - page 135

7
Lessons Learned
The application examples presented in Part II (Chapters 3 - 6) demonstrate
that in silico experiments of different flavors can be realized with the Bio-
jETI framework. They cover a broad range of typical bioinformatics work-
flow scenarios, concerned with different thematic areas, different software
components, different service technologies, and also workflows of different
complexity. They illustrate how the constraint-driven workflow develop-
ment methodology helps mastering the manifold workflow variants and
how PROPHETS' ability to flexibly formulate domain-specific and problem-
specific constraints supports the workflow development process. As such, the
application scenarios provide on the one hand qualitative evidence of the
method's applicability on real-world application scenarios, and on the other
hand they are suitable as a proper basis for further considerations about
the applied workflow development technology. In this context, this chapter
focuses specifically on the “lessons learned” regarding the constraint-driven
workflow design methodology that has been introduced to the Bio-jETI frame-
work in the scope of the work underlying this topic.
Section 7.1 discusses the evaluation of particular key figures of the synthe-
sis process in order to quantify the capabilities and limitations of the synthesis
framework systematically. In fact, the applied evaluation scheme is suitable to
show how effectively the constraint-driven method can guide the automated
search for workflows according to the user's (high-level) intents, but that its
performance is impacted by the state explosion of the search space that the
synthesis algorithm faces sooner or later. Interestingly, however, constraints
are also a key means with regard to dealing with the state explosion problem:
They reduce the size of the search space as they narrow the set of admissible
solutions. Thus, they do not only guide the search towards the actually in-
tended solutions, but at the same time help to delay performance problems
caused by state explosion.
Section 7.2 then summarizes the experiences gained during working on
the application scenarios and the results from the evaluation. The resulting
loose programming pragmatics provide guidance for domain preparation and