demonstrating its superior performance. Aertset al.( 2006 ) perform latedata
integration (for classification of data integration approaches, see Sect.4.1)of
multiple heterogeneous data sources: they generate distinct prioritization scores
for each data source separately, which are then fused into a global ranking using
order statistics. Another study uses Bayesian framework forearlydata integration
to integrate 16 distinct genomic data features and construct the functional-linkage
network (FLN) of human genes.^39 FLN is further used to prioritize candidate genes
based on network closeness to known disease-associated genes.
3 Biological Data and Their Network Representation
The human cell is a molecular factory where biomolecules rarely perform their
function in isolation, but rather they cooperate with other molecules to provide
specific functions. In the interdisciplinary field ofsystems biology, which deals with
the emergent properties of interacting molecules in an organism, networks are very
useful mathematical concept for representing and modeling different types of
molecular interaction. A network (graph), denoted asG¼(V,E), comprises two
sets: a set of vertices (nodes),V, and a set of edges (links),E, connecting pairs of
nodes.^40 A node can represent a biological molecule, e.g., a gene, a protein, a
metabolite, an RNA. We can also observe nodes at the phenotype level, where they
can represent diseases, or drugs. An edge can represent a physical binding (e.g.,
between a pair of proteins), a functional association (e.g., between a pair of genes),
or various types of similarities between nodes (e.g., a chemical similarity between
drugs). Depending on the type of an interaction, we can distinguish between
directedandundirectednetworks. For instance, a binding between two proteins is
usually represented as an undirected edge, while a chemical reaction that converts
one metabolite into another is usually represented as a directed edge.
Depending on the nodes’local connectivity, many biological networks have
heterogeneous, orscale-freearchitecture, meaning that these networks are charac-
terized by a small number of highly connected nodes (hubs), while most nodes
interact with very few other nodes.^41 This structural network property indicates that
molecular networks do not emerge in a completely random manner, but quite
opposite, they are a result of very complex processes taking place inside the cell.
Scale-free and geometric networks have been considered as null models for deriv-
ing further conclusions using these networks.^42
The ever-increasing accumulation of interaction data has made a construction of
molecular networks possible (see Table 1 ). The network perspective allows us to
(^39) Linghu et al. ( 2009 ).
(^40) West ( 2000 ).
(^41) Albert ( 2005 ).
(^42) Barabasi and Oltvai ( 2004 ), Higham et al. ( 2008 ), and Pržulj et al. ( 2004 ).
146 V. Gligorijevic ́and N. Pržulj