Biology Reference
In-Depth Information
more classical physiological disciplines, Cummins-style functional analysis has
been performed successfully. Applied to large networks, however, the structuring
becomes somewhat problematic because as a result of the distributed nature of
the functions in these cases, the limits of a functional subunit are not necessarily
sharp. Instead of following structural borders that may exist in the network,
functional analysis cuts the network into modules according to the functions
the descriptions (the analytical account) can address. The position of these cuts,
however, depends on how functional contributions can be conceptualized. The
cuts need not coincide with the actual structure of a network and in most cases
the functional modules will not be as independent from each other as they appear
under this decomposition. Functional decomposition may consequently distort
the picture of the network.
The question is, then, whether functional decomposition is an adequate
heuristic for the decomposition of large-scale networks in cases where the goal
is a faithful representation of a whole network in a model. The problem of
decomposing a network faithfully is claimed to have been solved by various
formal methods for obtaining 'unbiased modules' (see references in Section 6.2).
These methods do not rely on a functionality criterion but refer to structural and
dynamical properties of networks instead. Here, networks are viewed as 'nearly
decomposable' into structurally and dynamically largely autonomous compo-
nents or modules sensu Herbert Simon (1969), providing 'natural joints' along
which a network can be decomposed to yield modules. We note, however, and
again in the vein of Simon's pragmatic 'bounded rationality' assumption (“any
real-world system is too complex to study in all of its complexity, so we must
make simplifications”), that there are no good reasons to expect these methods
to yield a (unique) 'best' decomposition, whatever that should be; the 'best' one
is likely to get its robust results based on triangulation, which will always reflect
the researchers' particular aims with the modeling enterprise at hand (see, e.g.,
Wimsatt, 1981; Simon, 1982; cf. also Section 8.3).
13
,
14
The structural/dynamical decomposition strategy is supposed to give a less
distorted picture of a network than (often 'intuitive') functional decomposition
(Bechtel and Richardson, 1993; Schaffner, 1998; Papin et al. 2004).
15
Rather
than pursuing the question whether the various methods for 'unbiased modular-
ization' that have been proposed may deliver the goods, we here want to point
therefore depend on the concepts by which the contributions to a capacity are classified. This relativization,
which could in principle be almost arbitrary, may be overcome, however, by referring to the ontogenetically
effective fixation of components of functionally organized systems (Krohs, 2004 and submitted).
13
See Callebaut and Rasskin-Gutman (2005) for state-of-the-art applications of near-decomposability to com-
plex biological (and other) systems and discussion of the epistemological issues they raise.
14
Introducing an evolutionary perspective may also help structuring a network in cases where evolution results
in recurring, dynamic organizational principles (de Atauri et al., 2004).
15
Ideally for the modeler, of course, the two sets of modules arrived at on the different routes would coincide,
but in general they will not.
