examine the structure of these networks by applying graph theoretic approaches and
to further bridge the structural characteristics with the related biological function.
Some types of data contain information on the relationships between different types
of biological entities, such as the relations between drugs and diseases or between
genes and diseases. These types of data are more suitable for abipartitenetwork
representation. Abipartite graphis denoted asG¼(U,V,E), whereUandVare
disjoint sets of nodes (partitions) andEis the set of edges such that every edge in
Econnects a node inUto a node inV.^43 In this section, we do not describe
computational tools or commonly used algorithms from graph theory for the
analyses of network data. An interested reader may find more details on this topic
in Newman ( 2010 ) and Pržulj et al. ( 2004 ).^44 Instead, we comprehensively sum-
marize all available data for biomolecules, diseases, and drugs frequently used in
many studies in data integration, and we provide detailed instruction on how to
construct individual networks. We mainly focus on the following types of biolog-
ical networks: protein-protein interaction, metabolic, gene coexpression, genetic
interaction, signal transduction, gene-disease bipartite network, drug-target bipar-
tite network, drug interaction network, and directed acyclic networks representing
ontologies (gene or disease ontology).
3.1 Biological Networks
3.1.1 Protein-Protein Interaction (PPI) Networks
As mentioned earlier, PPI networks represent physical interactions (bindings)
between proteins in a cell. Proteins are represented as nodes, while their mutual
physical bindings are represented as undirected edges. Many previous studies have
dealt with the analysis and interpretation of these networks in many different
organisms.^45 Two large-scale, most widely used experimental techniques for
extraction of protein-protein interactions are yeast-2-hybrid (Y2H) screening^46
and affinity purification methods followed by Mass Spectrometry (AP/MS).^47
These experiments generate data with many flaws. For example, Y2H experiments
generate many false positive interactions; MS experiments are not initially designed
to identify binary protein interaction, but instead they identify protein complexes,
and the main problem is extraction of binary interactions out of these complexes. In
many data integration studies, the following data repositories of PPI networks have
(^43) West ( 2000 ).
(^44) Newman ( 2010 ) and Pržulj et al. ( 2004 ).
(^45) Barabasi and Oltvai ( 2004 ), Pržulj ( 2011 ), and Stelzl et al. ( 2005 ).
(^46) Consortium AIM ( 2011 ), Dreze et al. ( 2010 ), Giot et al. ( 2003 ), Ito et al. ( 2000 ), Li et al. ( 2004 ),
Stelzl et al. ( 2005 ), and Uetz et al. ( 2000 ).
(^47) Gavin et al. ( 2006 ) and Krogan et al. ( 2006 ).
Computational Methods for Integration of Biological Data 147