PSB-Tutorial

Informations

• Publié par: Idru
• Nombre de lectures: 25
• Langue: English

Extrait

DYNAMICS OF BIOLOGICAL NETWORKS:
OUTLINE OF SESSION TUTORIAL
*TANYA Y. BERGER-WOLF
Department of Computer Science, University of Illinois at Chicago
Chicago IL 60607, USA
†TERESA M. PRZYTYCKA
National Center of Biotechnology Information, NLM, NIH
Bethesda MD 20814, USA
‡MONA SINGH
Department of Computer Science, Lewis Sigler Institute for Integrative Genomics
Princeton University, Princeton NJ 08544, USA
#DONNA K. SLONIM
Department of Computer Science, Tufts University and Department of Pathology, Tufts University School of Medicine
Medford, MA 02155


1. Introduction

Network analysis provides a unifying language to describe relations within complex systems and
has played an increasingly important role in understanding biological systems. Over the past
decade, high-throughput experimental and computational methods have been developed to infer
and predict the structure of gene and protein networks. As a result, large-scale cellular networks
have been obtained for a wide range of organisms across the evolutionary spectrum.
Computational analyses of these networks have great potential in uncovering cellular
organization, pathways and functioning. Initial analyses of these networks have focused on
uncovering general topological features, and the majority of current approaches have treated these
networks as static and unchanging. Yet, most biological networks change temporally, spatially
and in a context-dependent manner. Therefore, in addition to a description of these networks as
collections of nodes and edges, researchers have began to elucidate dynamic properties of
biological networks. In molecular networks, this is frequently obtained by integrating static
interactions (such as protein-protein or regulatory interactions) with time- or environment-
dependent expression data, protein localization data, or other contextual information.

Dynamical properties of biomolecular networks can be viewed from a wide spectrum of
perspectives, starting from analyzing the changes in network properties over time or space, to
beginning to decipher signaling and regulatory pathways, to more detailed quantitative modeling
of metabolic and signaling pathways using continuous time ODE model representations. In this
tutorial, we focus on new approaches for analyzing network dynamics that have emerged in
response to the emergence of various types of large-scale experimental data rather than on more
established methods for quantitative modeling of network kinetics. Additionally, while there is a

* Work partially supported by NSF CAREER grant IIS-0747369 and NSF grants IIS-0705822 and IIS-0612044
† Supported by the Intramural Research Program of the National Institutes of Health and National Library of Medicine
‡ Work partially supported by NSF grant CCF-0542187, NIH grant GM076275 and NIH Center of Excellence grant P50 GM071508
# Supported in part by NIH grants LM009411 and HD058880 growing body of work on the challenging problem of inferring regulatory networks and pathways
from (primarily) gene expression data, due to time and space constraints, we limit our discussion
to approaches that largely assume that at least a partial, static snapshot of the network is
available (e.g., consisting of physical, phosphorylation or regulatory interactions), and that can
then be analyzed, perhaps in conjunction with other types of high-throughput data (e.g., eQTL or
gene expression data) to infer dynamic features and processes within these networks (e.g.,
pathways, modules, and varying topological features).

The tutorial is organized as follow. We start by briefly reviewing the available types of high-
throughput interaction data comprising cellular networks, as well as the other types of data that are
typically used in conjunction with these for dynamic network analysis. Next, we will briefly
discuss methods to infer pathways form (primarily) static interaction data. Then, we outline how
the network scaffold can be augmented with other types of more dynamic data to better uncover
regulatory networks and pathways. Finally, we describe general graph-theoretic analyses
techniques when dynamic networks are known, as well as provide discussion of how the basic
graph representation is changed.

2. Determination of Interaction Networks and Basic Static Analysis

Proteome-scale physical interaction networks, or interactomes, have been determined for several
organisms, including yeast and human. In simple graph theoretic terms, these networks have
vertices for each protein (or other macromolecule of interest), and edges between proteins that
interact. Biological networks can be comprised of direct physical interactions between proteins
(typically obtained via two hybrid analysis (Fields and Song 1989)) as well as of interactions
indicating that two proteins are part of the same multi-protein complex (reviewed in (Bauer and
Kuster 2003). High-throughput experiments have also linked together proteins in several other
ways, and it is possible to build large-scale networks consisting of links between proteins that are
synthetic lethals (Tong, Evangelista et al. 2001) or are coexpressed (reviewed in (Lockhart and
Winzeler 2000), or between proteins where one regulates (Lee, Rinaldi et al. 2002) or
phosphorylates the other (Ptacek, Devgan et al. 2005) (reviewed in (Zhu, Gerstein et al. 2007). In
addition to interaction networks that have been determined experimentally, there are a number of
computational methods for building functional interaction networks, where two proteins are linked
if they are predicted to perform a shared biological task (reviewed in (Galperin and Koonin 2000).

Computational analyses of these networks have identified global topological features (Barabasi
and Oltvai 2004) and have revealed a modular organization (Hartwell, Hopfield et al. 1999) with
highly connected groups of proteins taking part in the same biological process or protein complex
(Rives and Galitski 2003; Spirin and Mirny 2003). Dozens of papers for analyzing protein
interaction networks have focused on clustering these networks in order to uncover
complexes or functional modules (e.g., see (Bader and Hogue 2003; Brun, Chevenet et al. 2003;
King, Przulj et al. 2004; Brohee and van Helden 2006) and the review (Sharan, Ulitsky et al.
2007)). Complexes and functional modules are typically identified as subnetworks that contain
nodes that more strongly connected the sub-network than with the rest of the network.
Additionally, there has been considerable progress in predicting biological processes from
interaction networks based on extending the concept of guilt-by-association, where proteins are
annotated by transferring the functions of the proteins with which they interact (e.g., see
(Schwikowski, Uetz et al. 2000; Hishigaki, Nakai et al. 2001; Deng, Zhang et al. 2003; Letovsky and Kasif 2003; Vazquez, Flammini et al. 2003; Karaoz, Murali et al. 2004; Nabieva, Jim et al.
2005)). Biological networks can be visualized via systems such as Osprey (Breitkreutz, Stark et al.
2003) and Cytoscape (Shannon, Markiel et al. 2003), with many interactive visualization and
analysis tools available as plug-ins for Cytoscape.

3. Connecting Network Properties to Cell Dynamics

As mentioned above, the first steps in the analysis of protein interaction networks included
identification of protein complexes, functional modules and pathways. Although these concepts
are inherently static, a more careful analysis of the corresponding sub-networks may provide
insight into ordering of events within a signaling pathway or formation of mutiprotein complex.
Steffen et al. were first to develop a computational approach for generating static models of signal
transduction networks. Their approach utilizes protein-interaction maps generated from large-
scale two-hybrid screens and expression profiles from DNA microarrays (Steffen, Petti et al.
2002). In the heart of their approach was identification of relatively short pathways connecting
pairs of proteins. This was later refined by Scott et al. (Scott, Ideker et al. 2006).
Recently, Banks et al. (Banks, Nabieva et al. 2008) introduced network schemas as a more general
way to specify and search for signaling and regulatory pathways within interaction networks.
Zotenko et al. introduce a novel graph-theoretical framework that enables automatic construction
of a tree-like representation, that, depending on the nature of the network, is potentially capable of
elucidating temporal relations between functional groups (Zotenko, Guimaraes et al. 2006).

Another important line of research attempts to connect static properties of a network with the
dynamic behavior of the cell in response to perturbations. Jeong et al. were the first to observe
that that high-degree nodes in a protein interaction network tend to correspond to proteins that are
essential for survival of yeast cells in optimum conditions (Jeong, Mason et al. 2001). The relation
between protein essentiality and global and local topological features of the protein interaction
network has been broadly investigated by many authors (Yu, Greenbaum et al. 2004; Hahn and
Kern 2005; Batada, Hurst et al. 2006; He and Zhang 2006; Yu,