Lessons Learned - User-Level Workflow Design: A Bioinformatics Perspective - page 148

Among the application examples, this has become most evident for the
phylogenetics analysis workflow scenario discussed in Chapter 3: The first
workflow examples in this scenario (cf. Section 3.2) are based on the DDBJ
web services, which provide access to several standard sequence databases
and sequence analysis tools. Technically, the DDBJ web services are a (posi-
tive) exception among bioinformatics web services, as they provide extremely
simple service interfaces and are hence convenient to use for workflow ap-
plications. However, as they provide only a limited range of functionality,
the later, more comprehensive workflow examples of the phylogenetics sce-
nario are based on the EMBOSS tool suite. Principally, most EMBOSS tools
are readily available as web services, provided by the EBI's Soaplab project
[14, 15]. However, as many EBI web services, they provide rather complicated
interfaces (introducing loads of technical data types and requiring several ser-
vice calls for executing one tool, as the services are run asynchronously) and
are hence inconvenient to use. Consequently, the EMBOSS tools have simply
been provided as jETI services. Due to the availability of complete command
line interface descriptions in the ACD files (cf. Section 3.3), this tool inte-
gration process could even be completely automated. Also the other three
application scenarios described in this topic apply the jETI technology for
service provisioning.
Combining Services
Unless the adequate service interface corresponds directly to a single call to a
tool or API, this typically involves proper combination of several small steps
into a single, more abstract unit of functionality that then constitutes the
resulting service.
In the course of investigating further application areas of the Bio-jETI
framework as described in this topic, there have been different attempts to
provide systematic integrations of specific programming languages or APIs.
For instance, the jR plugin for the jABC framework, developed by Clemens
von Musil in his diploma thesis [334], provides means for automatic genera-
tion of SIBs for GNU R library functions and also for the execution of the
corresponding workflow via a (remote) R engine. The integration and exe-
cution facilities work properly, but as the granularity of the generated SIBs
essentially corresponds to single R statements, they are at the end of the day
simply too low-level for the application experts who are the actual target
users of the framework. Analogous experiences have been reported from simi-
lar experiments concerning a Matlab integration (Wolfgang Jansen, Potsdam
University, personal communication).
Consequently, for the integration of such functionality it is more adequate
to proceed as in the microarray data analysis scenario described in Chapter 6.
There, the R (or, more precisely, Bioconductor) functions that constitute the
“heart” of the envisaged services, are wrapped appropriately. Concretely, the
service does not call the library function directly, but a small script that also