Biology Reference
In-Depth Information
Clustering approaches on protein networks do have their limitations,
however. They can usually assign a given protein to one cluster only,
whereas in reality a protein may function as part of several distinct
groups. These distinct groups could be variants of a certain protein com-
plex, or even two completely different functional contexts in which a pro-
tein can act, a phenomenon termed “moonlighting”.
32
For clustering
approaches, such multiple memberships in distinct groups can cause
problems and are usually not reproduced in the clustering results. One
approach to overcome this is to iteratively execute multiple different clus-
tering runs, using various parameters, thereby generating an ensemble of
clusters that can then be filtered and analyzed for proteins present in dis-
tinct sets of clusters. This approach has recently been applied to the entire
yeast proteome,
33
and has revealed that many protein complexes indeed
show considerable variations in their makeup, but that small groups of
proteins nevertheless form “cores” or “modules” (subcomplexes) which
are quite stable in their composition.
Such small cores of interacting partners (often consisting of two or
three proteins only) have also been termed “motifs”, especially in studies
that focus on their topological arrangement in the network.
34,35
To define
a motif, consider that even a small number of nodes can be connected in
many different topologies: they can have a minimal number of connec-
tions only so that their interaction topology can be stretched out like
beads on a string, or they can be fully connected so that each protein is
linked with each of the others. Each such configuration is one type of
motif. The number of possible connection topologies is particularly high
in the case of directed networks such as networks of transcriptional reg-
ulatory interactions, where it makes a difference whether protein A reg-
ulates protein B or vice versa. In these types of networks, motifs have
been described first. Intriguingly, in actual networks derived from exper-
imental data, the various possible types of motifs are not all equally fre-
quent — some are far more abundant than they should be under a
random expectation. This is taken as evidence that these modules form
distinct regulatory circuits, for example, feedback loops, feedforward
loops, switches, delay elements, and so on. Indeed, for many of these
motifs, their predicted role and quantitative behavior have been con-
firmed experimentally (see, for instance, Refs. 36-38).