Phylogenetic Analysis Workflows - User-Level Workflow Design: A Bioinformatics Perspective - page 83

are extremely large and broad and therefore not reasonably representable
on normal paper page size. Hence, the first solution graph shown in the
figure is the graph that represents the 132 solutions that are obtained when
refinements 1, 2a and 2b are applied. As it is still extremely broad, is it
shown as a whole at the top of the figure, and additionally its two particularly
interesting parts are shown somewhat enlarged below. The first one comprises
the start state (shaded in gray) and some of the states in search depths 1 and
2. It is clearly visible that there are already quite a number of possibilities for
the first step from the start state, and then extremely many for the second.
As visible in the second enlarged part, of the graph, which comprises an
accepting state of the solution graph in depth 3, less possibilities remain for
the third step. The graphs for the solution sets for the last three refinement
steps, shown at the bottom of the figure, are then indeed extremely small,
only representing two or one solutions.

Refinement 1: Exclusion of Useless Services

When taking a look at the actual solutions, it is striking that several of the
solutions that are obtained when no constraints are applied contain services
that are useless in the sense that they distract from the actual synthesis
problem. Such services are, for instance, services that require no input but
produce new data that is planted into the workflow, like the makenucseq and
makeprotseq services that create sequence data “from scratch”. To avoid
the inclusion of such services in the solutions, a constraint can be defined
that excludes all services that do not require any inputs and thus may cre-
ate superfluous data. Therefore, the service ontology has been extended by
an auxiliary ServiceWithNoInput class, under which all these services were
grouped, and then the exclude service constraint (cf. Section 2.3.2) could
simply be used:

Do not use ServiceWithNoInput .
As Table 3.2 (row 2) shows, this constraint decreases the number of solutions
that are found considerably: the two one-step solutions remain, but only 193
solutions are found in depth 2, then 10,130 in depth 3, and 385,518 in depth 4.

•

Refinement 2: Exclusion of Unwanted Services

Another observation about the solutions is that some of the workflows contain
services that do in fact perform operations on the available data, but which
are actually unwanted within the intended solution. For the present example,
for instance, it can be assumed that the input sequence data is suitable for
further analysis and does not have to be edited beforehand. Accordingly,
Sequence editing services (like, e.g., degapseq ) can explicitly be excluded
from the solutions, again using the exclude service constraint:

• Do not use Sequence editing .