Computing Pathways to Systems Biology: Key

Contributions of Computational Methods in Pathway

Identification

Sriganesh Srihari, Mark A. Ragan

arXiv (arXiv: 1304.5565v1)

Generated on May 06, 2025

 Computing Pathways to Systems Biology: Key
Contributions of Computational Methods in Pathway
Identification

Abstract
Understanding large molecular networks consisting of entities such as genes, proteins or RNAs that
interact in complex ways to drive the cellular machinery has been an active focus of systems biology.
Computational approaches have played a key role in systems biology by complementing theoretical
and experimental approaches. Here we roadmap some key contributions of computational methods
developed over the last decade in the reconstruction of biological pathways. We position these
contributions in a 'systems biology perspective' to reemphasize their roles in unraveling cellular
mechanisms and to understand 'systems biology diseases' including cancer.

COMPUTING PATHWAYS TO SYSTEMS BIOLOGY: KEY CONTRIBUTIONS OF COMPUTATIONAL

METHODS IN PATHWAY IDENTIFICATION Sriganesh Srihari and Mark A. Ragan The University of
Queensland, Institute for Molecular Bioscience, St. Lucia, QLD 4072 Australia. Abstract Unde rstanding
large molecular networks consisting of entities such as genes, proteins or RNAs that interact in
complex ways to drive the cellular machinery has been an active focus of sy stems biology.
Computational approaches have played a key role in systems biology by complementing theoretical
and experimental approaches . Here we roadmap some key contributions of computational methods
developed over the last decade in the reconstruction of biological pathway s. We position the se
contributions in a ‘systems bio logy perspective’ to reemphasize the ir roles in unravelling cellular
mechanisms and to understand ‘systems biology diseases’ including cancer. 1. Introduction To
understand the fun ctional organis ation of cell s or higher biological units , often it is benefic ial to
conceptualize them as systems of interacting entities. For such a systems -level description, one needs
to know (a) the entities (“parts list”) that constitute the system , (b) the interactions among these entities
, and (c) their dynamic behavio ur und er changes to internal and external conditions [1]. The goal of
systems biology is to combine all this information into model s that capture current knowledge and
provide new insights and predictions about the system under previously unstudied conditions [2]. Early
attempts at systems biology suffered from inadequate data t o build reliable models and formulate
hypotheses; however, the recent advent of high -throughput technologies has brought the initial wave
of data to revive systems -level mode lling and anal ysis, thereby seeding a revolutionary change in
how biology is being studied and understood. Understanding of complex molecular networks consisting
of entities such as genes, proteins or RNAs connected by interactions for regulation or synthesis in cell
ular decision -making and responses has become a key focus of systems -level studies. Efficient
computational approaches can complement theoretical and experimental approaches to model,
analyse and distil knowledge from high -throughput data. The reconstructi on of biological pathways
through which cellular entities interact , signal and regulate cellular processes is certainly one of the
fundamental building blocks towards understanding whole biological networks. 1.1 Understanding
computational methods in a syste ms biology setting In order to decipher the correct and complete
picture of cellular organis ation, it is imperative to assess the entire collection of pathways as a whole
rather than individually ; the system as a whole has emergent properties that are n ot visible at the parts
level [3]. However the reconstruction of all pathways , together with their intricate network of cross - talk
and feedback loops , is a daunting task. Since pathways do not have definite start and end points or
distinct boundaries, modeli ng them computationally is a significant challenge. Nevertheless, all
computational methods developed to date model pathways as definite computable structures such as
paths, trees or subnetworks. Under these circumstances, it becomes all the more crucial t o place these
computational methods and their contributions in a systems biology setting to assess where we stand in
achieving this higher goal of systems -level understanding of cellular organi sation. In this article, we
roadmap some of the key computatio nal methods dev ised over the last decade for biological pathway
reconstruction from high -throughput data . Although we mainly look at methods reconstructing
regulatory or signalling pathways , this illustrates how computational methods have contributed to this
area more broadly . While some methods have focused on identifying general pathways, others have
specifically considered
dysregulated and disease pathways, while still others have looked at additive, alternative or
compensatory relationships among pathwa ys. Further, as these methods evolved, so did their
mechanisms to integrate diverse “omics” data , mainly genomics (gene expression, methylation,
mutation, regulation , genetic interaction) and proteomics (protein interaction). It is both imperative and
interesting to put all these developments together and consider where we stan d in deciphering
systems -level molecular network s and evaluate its implications. 2. General pathway identification In
an elegant study [4] conducted as early as 2001, Ideker et al. concentrated on a core pathway, GAL
(galactose utilization) , in the yeast Saccharomyces cerevisiae , implement ing an integrated approach
involving molecular expression and interactions to understand how its genes are regulat ed. After
assembling the GAL path way based on available information, these authors systematically perturbed
each gene and measured the response through expression of a global gene set. Around the same time,
large -scale protein -protein (PPI) and protein -DNA interactions were being catalogu ed for yeast [5],
and this enabled Ideker et al. to assemble a global network of ~3000 interactions. This network was
used to identify paths connecting perturbed GAL gene s to every other affected gene. They identified
nine genes involved in glycogen accumu lation and protein metabolism , and several of unknown
function , that respon ded strongly to galactose induction. As more high -throughput protein interaction
[5-7] and gene expression data [8] began to appear, Steffen et al. [9] in 2002 took a similar appro ach,
assembling a large PPI network to draw possible linear paths of specified lengths starting at membrane
proteins and ending on DNA -binding proteins. They scored these paths using co -expression among
adjacent proteins to identify high -ranking paths corr esponding to known pathways. This method,
called NetSearch, enabled them to reconstruct the yeast MAPK (mitogen -activated protein kinases)
pathways involved in pheromone response, filamentous growth, and maintenance
of cell wall integrity. Similarly , Liu e t al. [10] identified candidate set s of proteins in a PPI network, and
inferred the highest -scoring order among the m by measuring gene co - expression among adjacent
proteins in each permutation of the set. With this scoring, they could identify the correct ordering among
proteins that corresponded to MAPK pathways in yeast. In a separate attempt , Friedman et al. [ 11,12]
used a Bayesian framework to learn expression profiles of genes perturbed in yeast mutants and used
it to infer pairwise expression correlat ion among proteins . Through this they assemble d pathways
involved in purine biosynthesis and non -homologous DNA double strand break repair. Alongside yeast,
large -scale PPI data from prokaryotes including Helicobacter pylori [13] and eukaryotes including
Caenorhabditis elegans [14] began to appear around 2001 -2002, which enabled researchers to search
in networks for pathways conserved across species. In a seminal work of this kind, Kelley et al. [15] in
2003 devised PATHBLAST, an efficient tool to align paths across multiple PPI networks. This tool
enables search for homologous paths across networks by accommodating “gaps ” and “mismatches” .
Th ey identifi ed 150 homologous paths of lengths ≥4 among the three species. Further, by self
-matching the yeast networ k, Kelley et al. identif ied about 300 paralogous paths that they grouped into
sever al functional pathway s. Following this success, Shlomi et al. [16] proposed QPath, which
improved on the results of PATHBLAST. In the meantime, as reports [17,18 ] of high f alse-positive
rates in high -throughput experiments began to surface, it became necessary to assess reliabilit y of
interactions before employing them in focused studies such as pathway identification. In a seminal
study combining reliability scoring and pat hway identification, Scott et al. [19] (2006) assigned a score
to every interaction in the network by combining three criteria: (a) the number of times a protein pair
was seen interacting in multiple experiments , (b) the Pearson correlation between express ion profiles
of the proteins , and (c) the ir small -world clustering coefficient. Using the resultant scored network,
they devised two algorithms to identify pathway structures. The first identified high -scoring simple
paths, while the second identified more -general st ructures including rooted trees and ‘series -parallel’
graphs. Through exp eriments on a yeast network of ~4500 proteins and ~14500 interactions, they
successfully reconstructed several pheromone -respon se pathways with high accuracy .
Aided by com putational tools and experimental approaches, the growth of public databases including
KEGG [20] for pathways and FunCat [21] and Gene Ontology (GO) [22] for functional annotations
enabled pathway identification methods to use these annotations for both pr ediction and validation of
results. Among the first to use such diverse data was PathFinder by Bebek et al. [23] in 2007. These
authors collected functional annotations from FunCat and GO for proteins in KEGG to build functional
templates for pathways. They then used these templates to mine pathways from the yeast PPI network,
and the high -support pathways were identified and scored using association rules mining [24].
PathFinder showed significantly better accuracy and sensi tivity compared to most earlier methods ,
and identif ied several missing links among proteins in annotated pathways in databases . In an attempt
to identify general substructures beyond linear paths, Zhao et al. [25] in 2008 devised an integer linear
programming (ILP) -based approach to mi ne the yeast PPI network. They modeled pathways as
compact subnetworks between fixed starting and ending points. These subnetworks were scored using
reliability scores on the edges : linear paths were scored as the sum of the edges in the paths, while
gener al subnetworks were scored as th e sum of the constituent edges. An ILP -based model wa s
then proposed to extract high - scoring subnetworks from the PPI network. Experiments on
subnetworks identified between membrane proteins and transcription factors showed that many
pherom one-response pathways and signalling pathways for filamentous growth were re constructed
with high accuracy . Liu et al. [26] (2009) noticed that most signals between proteins (for example
activation, inhibition, phosphorylation, dephosphor ylation and ubiqui tination) were directional, so
identifying the correct direction of interactions among proteins was crucial for accurate reconstruction of
signalling pathways. They proposed a signal -flow model to orient interactions in the PPI network, f or
which they used domain interaction information among proteins. Based on the deduced orientations,
they identif ied potential upstream - downstream relationships within protein pairs. This method
successfully reconstructed several signalling pathways from t he human PPI network, which matched
ones annotated in KEGG and other databases.
Independent to these approaches, Boolean [27,28], Petri Nets [29] and ordinary differential equations (
ODEs) [ 30] are a few other models that were proposed, but these have mainly focused on simulation
and study of behavior of known pathways as against pathway inference from high -throughput datasets.
3. Exploring relationships among pathways As methods to identify pathways improved, it became
interesting to understand the relationships among different pathways and ho w they constituted the
larger “pathway network” to regulate and govern cellular processes. This was further fuel led by the
reali sation that many diseases, including cancer, arise from a complex interplay between pathways
acting in additive, compensatory or alternative ways to maintain aberrant behavio ur of cells . While the
genomics and proteomics data had already proven useful, the availability of genetic -interaction data
from systematic knock -out experiments in yeast and other organisms [31-33] further aided these
studies. Among the seminal works in this direction , Kelley and Ideker [ 34] (2005 ) combined PPI and
genetic -interaction (GI) networks to understand pathway relationships. They proposed that several
pathway s linking proteins in the PPI network were related by between -pathway interactions in the GI
network . This between -pathway model (BPM) proposed that such pathways are involved in
compensatory functions and buffered the loss of one another. These pathways f orm alternative or
redundant functional groups to maintain the robustness of pathway mechanisms. BPM prompted
further work to look at PPI and GI interactions in an integrated manner to decipher such pathway
relationships. One of the early works that took this forward was by Ulitsky and Shamir [35] (2007) , who
assembled a large network combining PPI and GI data,
and systematically searched for BPM structures. They found that several of the pathways in KEGG [28]
were related by BPM structures, indicating tha t these pathways functioned in a compensatory fashion.
BPM relationships between proteins in different complexes with ‘pivot’ or shared proteins among the
complexes were essential to all the host complexes. Subsequently Hescott et al. [36] (2009) integrat ed
gene -expression data to evaluate BPM structures identified from PPI and GI networks in order to
further refine re dundant pathway identification. Figure 1: Models of synthetic lethality relationships
seen between proteins within and betwe en pathways and complexes (adapted from Le Meur and
Gentleman [ 37] with permission from Genome Biology ). In this context, a subset of genetic
interactions called synthetic lethality (SL) interactions have gained immense interest due to their
prominence in connecting compensatory pathways and functions. Le Meur and Gentleman (2008) [37]
analysed the enrichment of SL interactions within and between complexes and pathways ; most SL
interactions were between -pathway and -complexes, while a considerable number were also within
these structures (see Figure 1 ). These within -complex and -pathway interactions ensured internal
robustness to these structures by buffering the functions of proteins. In separate work, Ma et al. (2008)
[38] proposed that finding bipartite connected subnetworks or bicliques in SL
networks can help to identify groups of proteins belonging to redundant pathways. To this end, they
searched for bicliques among SL interactions, and were able to identify several interesting relationships
between pathways governing functions including DNA repair and DNA replication, and tubulin folding
and mitosis. Brady et al. (2009) [39] further extended this approach to search for structures that they
called stable bipartite subgraphs, and identified several new pairs of redundant pathway relationships.
4. Pathways in dysregulated functions and diseases Many diseases , including cancer , result from
dysregulated pathways and their complex interactions , and it is becoming increasingly clear that
mapping these pathways is crucial to fully understand these complex diseases. This requires
integrating diverse information from gene exp ression, mutation and regulation , protein interaction
datasets and others; this has prompted integrative or “multi -omics” research towar ds pathway
identification. Early work by Schadt et al. (2005) [40] and Tu et al. (2006) [41] focused on identifying
possible linkages between (causal) mutations in DNA sequences or genes , and differentially
expressed (target) genes under disease conditions . The ir aim was to trace paths of differential
expression from the targets back to the causal genes ; these paths were hypothesised to constitute
dysregulated pathways in the disease. Both groups approached this by looking for paths between
target and causa l genes through PPI networks. The interactions in the network were scored based on
the correlation between each gene in the network and the target genes. Tu et al. simulated random
walks from target genes, and the most frequently visited causal genes were evaluated for possible
mutational characteristics, while the corresponding paths were analysed for involvement in disease
mechanisms. Following these approaches, several methods were devised to simulate flows to identify
paths between causal and target ge nes. An important method by Suthram et al. [42] in 2008 used
electric -circuit -based mode lling [43] to simulate flow s in the network. The motivation
was that the random -walk -based methods were stochastic and required many simulations (about
10000 times in T u et al.) to determine the causal gene. To propose a deterministic steady -state
solution, Suthram et al. equated these random walks to the flow of electric current , and solved the
network using electric -circuit theory ( Kirchhoff's and Ohm's Laws ). Equating the amount of current
flow through each node and edge in the network to the expression level and importance of the genes,
they could determine the ‘true’ causal gene s with high accurac y. The electric circuit method was
successfully adopted by Kim et al. (2009) [44] in the study of glioblastoma multiforma. They selected a
set of differentially expressed target genes that covered 158 glioblastoma cases , and then i dentified
possible genomic loci harbouring causal genes responsible for the differential expres sion of target
genes. By overlaying a human PPI network, Kim et al. found p robable paths from target to causal
genes for which they used the electric -circuit model of Suthram et al. The causal genes identified were
then evaluated for the disease cases they covered. Gene Ontology -based analysis of the identified
pathways showed high enrichment of processes involve d in glioblastoma. Following this success, He
et al. [45] used a similar approach to identify dysfunctional genes and modules in congenital heart d
isease (CHD). 5. Role of computational methods in future breakthroughs Among the kinds of genetic
interactions, synthetic lethality (SL) describes a scenario in which single -gene defects are compatible
with cell viability but a combination of gene defec ts results in cell death [46]. In essence , these SL
interactions provide functional buffering , sometimes de scribed as genetic canali sation , that is,
buffering of pathways against the tendency of new alleles or mutated genes to make non -optimal
phenotypes [47]. With the assembling of major pathways involved in core cellular processes affected in
cancer , including DNA replication, DNA damage repair and cell-cycle checkpoint s, it is now
clear that buffering among pathways maint ains viability in cancerous cells , potentiall y weakening a
nti-cancer therapies aimed at blocking individual pathways. However on the positiv e side, this hints
that “sweet spots” capable of overcoming such compensatory arrangements can be identified that may
lead to effective anti-cancer s trategies in the future [48]. The PARP/BRCA relationship has become the
poster child for SL -based cancer research today [48,49]. BRCA1 is a large protein expressed during
the S and G2 phases of the cell cycle, and involved in the DNA double -strand break (DSB) repair
pathway, one of several pathways required for maintaining DNA integrity. BRCA1 -mutant cells have a
defect in DNA damage repair, specifically in homologous recombination (HR) -mediated repair. In
normal cells, the loss of BRCA1 activity is sensed in S-phase , resulting in immediate TP53 -mediated
cell death. However, tumor cells acquire a state in which repeated transit through S -phase can be
accomplished despite loss of BRCA1 function. This leads to genomic instability and potential mutations
in ot her crucial ge nes including p 53 required to control cell proliferation. A major breakthrough in
BRCA1 -mutant cancers was heralded by the finding that BRCA1 mutant cells are sensitive to PARP
inhibitors. PARP1 is involved mainly in the DNA single -strand br eak (SSB) repair pathway. In the
context of PARP inhibition, unrepaired SSBs accumulate into DSB equivalents upon entry into S
-phase. In normal cells, these lesions are repaired by the HR -mediated DSB repair pathways. However
, in the absence of BRCA1 there may be greater reliance on PARP -mediated pathways, failing which
DNA DSBs accumulat e and lead to cell arrest and death. In essence , inhibition of PARP deals a
double blow to BRCA1 -deficient cells leading to cell death. Several PARP -inhibitor compounds are
now under clinical or pre -clinical trials for use in anti -cancer therapy based on this concept [48].
Although the PARP/BRCA relationship is only a conditional one and not a panacea, it reflects the extent
and kind of intricate relationship s that need to be deciphe red to map and understand the “pathway
network” , and therefore to understand diseases like cancer . As Laubenbacher et al. [3] rightly put it as,
“ cancer is a systems biology disease” . While our current knowledge of pathways in cancer is still inc
omplete, immense efforts are underway to identify new players (genes, proteins and whole pathways)
as well as to implicate existing
ones in new roles – for example, the recent (2009) implication of a SUMO -mediated pathway in the
BRCA1 response to genotoxic stress [50]. Computational methods have a key role to play alongside
experimental approaches. As an example, in recent remarkable research Rodriguez et al. [51] (2012)
constructed a Boolean network model of the Fanconi anaemia/ breast cancer (FA/BRCA) pa thway to
simulate the interstrand cross -links (ICL) repair process, whose inhibition is known to result in a
chromosomal instabil ity syndrome called F anconi an aemia. Rodriguez et al. mode lled knowledge of
ICL as logical rules , obtaining a Boolean network o f 28 nodes and 122 regulatory interactions. These
Boolean rules captured relationships among genes and pathways; for example, ICL can be responded
to either by generating a DSB that is subsequently repaired by the BRCA1 -mediated HR pathway , OR
by bypassing this with the help of proliferation cell nuclear antigen (PCNA) and translesion synthesis
(TLS) and repair ing it with the nuclear excision (NER) pathway. Next, by fixing the loss - and gain -of-
function mutants as 0 or 1 in the network, they performed dynam ical simulations to understand the
state of the network , that is, the alternative pathways favoured under various mutational conditions. In
this way they inferred key buffering mechanisms that may compensate for the defective FA/BRCA
pathway, which are wor th further research and experimental validation. 6. Conclusions In order to
decipher a correct and complete picture of cellular organi sation, it is necessary to take a systems -level
vi ew. Mapping the molecular network is a crucial step toward this goal. However, mapping the entire
network at one go can be daunting (and the cu rrent data are inadequate to support this) , so mapping
individual pathways is a rational way to proceed. Having said that, knowledge of individual pathways
must be reflected back to the systems -level context, lest we miss the bigger picture.
Computational approaches have played a key role in identification and mapping of pathways. In this
article, by drawing a roadmap of key contributions from computational approaches, and describin g
instances in which collective understanding of multiple pathways is necessary (for example, in cancer),
we have attempted to put all these developments in a systems -level perspective . Future
breakthroughs in understanding cellular mechanisms and diseases can come only by taking such a
systems -level view . To this end, computational approaches will continue to play a key role . Key
Points ■ Mapping the network of genes, proteins, RNA, and other molecules is a crucial step towards
realizing the ‘systems bio logy goal ’ of understanding cellular organi sation. ■ Computational methods
play a key role in this by complementing theoretical and experimental approaches. ■ We have
highlighted key contributions of computational approaches in id entification of signalling/ regulatory
pathways. These approaches have gone hand -in-hand with the improvements in high -throughput
techniques, and have integrated diverse “omics” datasets. ■ Through these contributions, we
reemphasize that computational methods have a key role in future breakthroughs in understanding
cellular organization and also complex systems -level diseases like cancer.
References [1] Kitano, H. Systems Biology: A brief overview. Science 2002; 295 :1662 --1664. [2]
Albert, R. Network inference, analysis, and modeling in systems biology. Plant Cell 2007; 19:3327
—3338 . [3] Laubenbacher, R., Hower, V., Jarrah, A., Torti, S.V., Shulaev, V., Mendes, P., Torti, M.,
Akman, S. A systems biology view of cancer. Biochim Biophys Acta 2009; 1796(2):129 —139. [4]
Ideker, T., Thorsson, V., Ranish, J., Christmas, R., Buhler, J., Eng, J.K., Bumgarner, R., Goodle tt, D.,
Aebersold, R., Hood, L. Integrated geno mic and proteomic analyses of a dystematica lly perturbed
metabolic network. Science 2001; 292(5518):929 --934. [5] Schwik owski, B., Uetz, P., Fields, S. A
network of protei n-protein interactions in yeast. Nature Biotechnol. 2000; 18:1257 --1261 . [6] Uetz, P.,
Giot, L., Cagney, G., Mansfield, T., Judson, R., Knight, J., Lokshon, D., Narayan, V., Srinivasan, M.,
Pochart, P. A comprehensive analysis of protein -protein interactions in Saccharomyces cerevisiae.
Nature 2000; 403:623 --627.
[7] Ito, T., Chiba, T., Ozawa, R., Yoshi da, M., Hattori, M., Sakaki, Y. A comprehensive two - hybrid
analysis to explor e the yeast protein i nteractome. Proc Natl Acad Sci 2001; 98:4569 -- 4574 . [8]
Hughes, T.R., Marton, M.J., Jones, A., Roberts, C.J., Stoughton, R., Armour, C., Bennett, H., Coffey,
E., Dai, H., He, Y. Functional discovery via a co mpendium of expression profiles. Cell 2000; 102:
109--126. [9] Steffen, M., Petti, A., Aach, J., D'ha eseleer, P., Church, G. Automated modelling of signal
transduction networks. BMC Bioin formatics 2002; 3:34 . [10] Liu, Y., Zhao, H. A computational
approach for ordering signal transduction pathway compo nents from genomics a nd proteomics data.
BMC Bioinformatics 2004; 5:158. [11] Friedman, N., Linial, M., Nachman, I., Pe’er, D. Using Bayesian
networks to analyse expression data. J Computational Biology 2000;7(3 -4):601 --620. [12] Pe’er, D.,
Regev, A., El idan, G., Friedman, N. Inferring subnetworks from perturbed expression profiles.
Bioinformatics 2001; 17(Suppl 1):215 --224. [13] Rain, J., Selig, L., de Reuse, H., Battaglia, V.,
Reverdy, C., Simon, C., Lenzen, G., Petel, F., Wojcik, J., Schachter, V., Ch emam a, Y., Labigne, A.,
Legrain, P. The protein -protein interaction map of Helicobacter pylori . Nature 2001; 409(6817):211
--215. [14] Boulton, S.J., Gartner, A., Reboul, J., Vaglio, P., Dyson, N., Hill, DE., Vidal, M. Combined
functional genomic maps of t he C. elegans DNA damage response. Science 2002; 295(5552):127
--131. [15] Kelley, B.P., Sharan, R., Karp, R.M., Sittler, T., Root , D., Stockwell, B., Ideker, T.
Conserved pathways within bacteria and yeast as revealed by global p rotein network alignment. Proc.
Natl. Acad. Sci. 2003; 100(20):11394 --11399 .
[16] Shlomi, T., Se gal, D., Ruppin, E., Sharan, R. QPath: a method for querying pathways in a prot
ein-protein interaction network. BMC Bio informatics 2006; 7:199 . [17] von Mering, C., Krause, R.,
Snel, B ., Cornell, M., Oli ver, S.G., Fields, S., Bork, P. Comparative assessment of large -scale data
sets of protein -protein interactions, Nature 2002; 417:399--403. [18] Bader, G., Chaudhuri, A.,
Rothberg, J., Chant, J. Comparative assessment of large -scale data sets of protein –protein
interactions. Nature Biotechnol. 2004; 22(1):78 --85. [19] Scott, J., Ide ker, T., Karp, R.M., Sharan, R.
Efficient algorithms for detecting signaling pathways in protein interaction networks. J Comp Biology
2006; 13(2):133 --144. [20] Kanehisa M., Goto S. KEGG: Kyoto Encyclopedia of Genes and Genomes.
Nucleic Acids Res 2000; 28:27 -30. [21] Ruepp A, Zollner A, Maier D, Albermann K, Hani J, Mokrejs M,
Tetko I, Guldener U, Mannhau pt G, Munsterkotter M, Mewes HW. The FunCat, a functio nal annotation
scheme for systematic classification of proteins from whole genomes. Nucleic Acids Res 2004;
32(18):5539 -5545 . [22] Ashburner M, Ball CA, Blake JA, Botstein D, Butler H, Cherry JM, Davis AP,
Dolinski K, Dwight SS, Eppig JT, Harris MA, Hill DP, Issel -Tarver L, Kasarskis A, Lewis S, Matese JC,
Richardson JE, R ingwald M, Rubin GM, Sherlock G. Gene ontology: tool for the unification of biology.
Nat Genetics 2000; 25:25 --29. [23] Bebek, G., Yang, J. PathFinder: mining signal transduction pathwa
y segments from protein -protein interaction networks. BMC Bioinformatics 2007; 8:335 . [24] Agarwal,
R., Srikanth, R. Fast algorithms for mining association rules. Proceedings of 20th International
Conference on Very Large Databases (VLDB) 1994; 487 --499.
[25] Zhao, X -M., W ang, R -S., Chen, L., Aihara, K. Uncovering signal transduction networks from high
-throughput dat a by integer linear programming. Nucleic Acids Res 2008; 36(9) :e48 . [26] Liu, W., Li,
D., Wan g, J., Xie, H., Zhu, Y., He, F. Proteome -wide p rediction of signal flow direction in protein
interaction networ ks based on interacting domains. Mol Cell Prot 2009; 8(9):2063 --3070 . [27]
Kauffman, S.A. Metabolic stability and epigenesis in randomly constructed gene sets. J. Theoretical
Biology 1969; 22 :437 --467. [28] Thomas, R., D’Ari, R. Biological feedback. Boca Raton, FL CRC
Press, 1990. [29] Peterson, J.L. Petri Nets. ACM J. Computing Surveys 1977; 9(3):223 --252. [30]
Hughey, J.J., Lee, T.K., Covert, M.W. Computational modeling of signalling netw orks. Wiley Interdiscip
Rev Syst Biol Med 2010; 2(2):194 -209. [31] Tong, AHY., Lesage, G., Bader, GD., Ding, H., Xu, H., et
al. Global mapping of the yeast genetic interaction network. Science 2004; 303:808 --813. [32] Boone,
C., Bussey, H., Andrews, BJ. Exploring genetic interactions and networks with yeast. Nat Rev Genet
2007; 8:437 --449. [33] Koh, J., Ding, H., Costanzo, M., Baryshnikova, A., Toufighi, K., Bader, G., Myer
s, C., Andrews, BJ., Boone, C. DRYGIN: a database of quantitative geneti c interact ion networks in
yeast. Nucleic Acids Res 2010; 38:D502 -D507 . [34] Kelley, R., Ideker, T. Systematic interpretation of
genetic inte ractions using protein networks. Nat Biotechnol 2005; 23(5):561 .
[35] Ulitsky, I., Shamir, R. Pathway redundancy and protein essentiality revealed in the Saccharomyces
cerevisiae interaction networks. Mol Sys Biol , 2007; 3:104 . [36] Hescott, B., Leise rson, M., Cowen, L.,
Slonim, D. Evaluating between -path way models with expression data. Proceedings of the 13th
International Co nference on Research in Computatio nal Molecular Biology (RECOMB) 2009; pp.372
--385. [37] Le Meur, N., Gentleman, R . Modeling synthetic lethality. Genome Biology 2008; 9:R135.
[38] Ma, X., Tar one, A.M., Li, W. Mapping genetically compensatory pathways fro m synthet ic lethal
interactions in yeast. PLoS ONE 2008; 3(4):e1992 . [39] Brady, A., Maxw ell, K., Daniels, N., Cowen, L.
Fault tolerance in protein interaction networks: stable bipartite s ubgraphs and redundant pathways.
PLoS ONE 2009; 4(4) :e5364 . [40] Schadt, E., Lamb, J., Yang, X., Zhu, J., Edwards, S., Guhathakurta,
D., Sieberts, S.K., Monks, S., Reitman, M., Zhang, C., et al. An integrative genomics approach to inder
causal associations between gene expression and disease. Nature Genetics 2005; 37:7 10—717. [41]
Tu, Z., Wang, L., Artb eitman, M.N., Chen, T., Sun, F. An integrative approach for causal gene
identification and ge ne regulatory pathway inference. Bioinformatics 2006; 22(14):e489 --e496 . [42]
Suthram, S., Beyer, A. Karp, R., Eldar, Y., Ide ker, T. eQED: an efficient method for interpreting eQTL
asso ciations using protein networks. Mol Sys Biol 2008; 4:162 . [43] Odum, E. Fundamentals of
ecology. W.B.Saunders, Philadelphia 1953. [44] Kim, Y -A., Wutchy, S., Przytycka, TM. Identifying
causal g enes and dyregulated pathways in complex diseases. PLoS Comp Bio 2009; 7(3):e1001095 .
[45] He, D., Liu, Z -P., Chen, L. Identification of dysfunctional modules and disease genes in congenital
heart dise ase by a network -based approach. BMC Ge nomics 2011; 12 :592 . [46] Ashworth, A., Lord,
C.J., Ries -Filho, J. Genetic interactions in c ancer progression and treatment. Cell 2011; 145. [47]
Hartma n, J., Garvik, B., Hartwell, L. Principles for the buffering of genetic variation. Science 2001;
291:1001 --1004. [48] Lord, C.J., Ashworth, A. The DNA damage response and cancer therapy, Nature
2012; 481:287 --294. [49] Amal, A., Ganesan , S. BRCA1, PARP, and 53BP1: conditional synthetic le
thality and synthetic viability. J Mol Cell Biol 2011; 3:66 --74. [50] Morris, JR. , Boutell, C., Keppler, M.,
Densham, R., Weekes, D., Alamshah, A., Butler, L., Galanty, Y., Pangon, L., Kuichi, T., Ng, T.,
Solomon, E. The SUMO modification pathway is involved in the BRC A1 response to genotoxic stress.
Nature 2009; 462(7275):886 —890. [51] Rodriguez, A., Sosa, D., Torres, L., Mo lina, B., Fras, S.,
Mendoza, L. A Boolean netwo rk model of the FA/BRCA pathway. Bioinformatics 2012; 28(6):858
--866.