Microarray Data Analysis Pipelines - User-Level Workflow Design: A Bioinformatics Perspective - page 130

in Section 6.2, some local services for the reading and writing of files and
for the loading of benchmark data sets are contained in the domain model.
Whereas the data loading/reading services require no inputs and the file
writing services produce no outputs, all other services in the domain require
one or more inputs and produce one or more outputs. Services that belong
to the same (bottom-level) group of the service taxonomy often also have
a similar input/output behavior. For instance, the Filtering services read
and write ExpressionSets, while most of the Preprocessing services create
ExpressionValues from AffymetrixCELData.

Constraints

The domain constraints in this example are used to describe the general over-
all structure of microarray data analysis pipelines, such as that DataLoading ,
Preprocessing and StatisticalAnalysis should be uses exactly once, and that
as last step a FileWriting service should be used in order to have results
stored to the local file system when the workflow finishes. Furthermore, do-
main constraints are used to express that the HTML-formatted results of
Annotation steps should be stored immediately, and that the currently avail-
able differential expression analysis service is not suitable for application on
the dilution benchmark data set. In terms of the template-based constraint
editor of PROPHETS, the corresponding constraints are:

• Enforce the use of DataLoading .

Enforce the use of Preprocessing .
Enforce the use of StatisticalAnalysis .

•

Do not use BenchmarkDataLoading more than once.
Do not use Preprocessing more than once.
Do not use StatisticalAnalysis more than once.

• Use FileWriting as last service in the solution.

• If Annotation is used, WriteHtmlFile has to be used next.

• If StatisticalAnalysis is used, do not use Filtering subsequently.

• At most one of LoadDilutionBenchmarkData and

DifferentialExpressionAnalysis4ReplicateArrays may exist.

These constraints describe the substantial, common characteristics of mi-
croarray analysis pipelines built with this domain model. During workflow
design, additional constraints can be used to guide the synthesis to partic-
ular solutions explicitly, for instance by demanding or avoiding the use of
specific services.

6.3.2 Exemplary Workflow Composition Problem

Microarray data analysis workflows are typically pipelines (i.e., linear se-
quences of analysis steps), and thus linear synthesis is suitable to create the