Biomedical Engineering Reference
In-Depth Information
Since the first such protein interaction maps became available, their struc-
tural and functional characterization have received great attention. Such net-
works reveal some simple structural features that they share with other
biological networks (20,32). In particular, it has been noted that the number d of
interactions per protein often resembles a power law, P ( d ) ~ d - H , where H is a
constant characteristic of the network. In addition, such networks show a high
degree of clustering of interactions within small groups of nodes. Such cluster-
ing is often measured through a clustering coefficient C (33). To define the clus-
tering coefficient C ( v ) of a node (protein) v in a graph, consider all k v nodes
adjacent to a node v , and count the number m of edges (protein interactions) that
exist among these k v nodes (not including edges connecting them to v ). The
maximally possible m is k v ( k v -1 )/2, in which case all m nodes are connected to
each other. Let C ( v ): = m /( k v ( k v - 1)/2). C ( v ) measures the "cliquishness" of the
neighborhood of v , i.e., what fraction of the nodes adjacent to v are also adjacent
to each other. In extension, the clustering coefficient C of the graph is defined as
the average of C ( v ) over all v . It can be orders of magnitude larger in biological
networks than in random networks of similar size and connectivity distribution.
Despite some intriguing propositions (2), the functional and biological signifi-
cance of such functional features is still unclear. In addition, because of their
size and complexity, these networks may harbor biologically important struc-
tural features that remain completely unexplored.
4.
MEDICAL APPLICATIONS
A hopefully temporary shortcoming of all presently available approaches to
identify protein interactions on a genome-wide scale is their high rate of detect-
ing spurious interactions and missing actually occurring interactions (7). Never-
theless, a number of commercial ventures (e.g., the German company Cellzome
[www.cellzome.com] or the French company Hybrigenics [www.hybrigenics.
com]) are already dedicating themselves to providing information on pairwise
protein interactions and protein complexes (protein networks in the strong sense)
to the pharmaceutical industry, for the purpose of drug discovery. Such com-
mercial interest is a strong indicator of the promise protein network information
holds for the future of drug discovery. However, the vast majority of studies
published to date used network information in the weak sense, as defined above,
in proof-of-principle biomedical applications. I will now provide a few exam-
ples of such applications.
4.1. Mechanisms of Drug Action
Analysis of protein expression can be useful in comparing the effects of
newly identified antibiotics to those of existing antibiotics. This serves the pur-
Search WWH ::




Custom Search