Journal of Computational Science 5 (2014) 332–344

Contents lists available at ScienceDirect

Journal of Computational Science

journal homepage: www.elsevier.com/locate/jocs

Using Boolean networks to model post-transcriptional regulation in

gene regulatory networks夽
Gianfranco Politano a,∗ , Alessandro Savino a,b , Alfredo Benso a , Stefano Di Carlo a ,
Hafeez Ur Rehman a , Alessandro Vasciaveo a,b
a
Politecnico di Torino, Control and Computer Engineering Department, Corso Duca degli Abruzzi 24, 10129 Torino, Italy
b
Consorzio Interuniversitario Nazionale per l’Informatica, Verres, AO, Italy

a r t i c l e i n f o a b s t r a c t

Article history: Gene regulatory networks (GRNs) model some of the mechanisms that regulate gene expression. Among
Received 15 January 2013 the computational approaches available to model and study GNRs, Boolean network (BN) emerged as very
Received in revised form successful to better understand both the structural and dynamical properties of GRNs. Nevertheless, the
27 September 2013
most widely used models based on BNs do not include any post-transcriptional regulation mechanism.
Accepted 4 October 2013
Since miRNAs have been proved to play an important regulatory role, in this paper we show how the
Available online 16 October 2013
post-transcriptional regulation mechanism mediated by miRNAs has been included in an enhanced BN-
based model. We resort to the miR-7 in two Drosophila cell fate determination networks to verify the
Keywords:
MiRNA
effectiveness of miRNAs modeling in BNs, by implementing it in our tool for the analysis of Boolean
Gene regulatory networks networks.
Post-transcriptional regulation © 2013 Elsevier B.V. All rights reserved.
Boolean networks
Complex systems
Network analysis

1. Introduction particular, researchers require automated formal modeling

techniques and appropriate analysis tools.
The genome of an organism plays a crucial role in regulating Several computational approaches have been proposed and
several cellular processes. With few exceptions, every cell in an developed in literature ([4] and further references therein) to model
organism contains the same genetic material. This implies that, gene regulatory networks (GRNs). They include partial/ordinary
after sequencing the whole genome of several living organisms differential equations [5,6], linear models [7], Bayesian networks
[1], and after knowing the functional meaning of thousands of pro- [8,9], Boolean networks [10,11] and Petri nets [12]. They can
teins encoded by these genomes [2], one of the principal remaining be divided into two groups: (1) discrete-state models and (2)
challenges is to understand the mechanisms that regulate gene continuous-state models. Discrete-state models assign a small num-
expression within the genome [3]. ber of discrete states to each node of the network, avoiding
The regulation of gene expression is achieved through genetic intermediate expression levels. Regulatory interactions between
regulatory systems organized in networks of interactions between nodes are described by logical functions. Bayesian networks,
genes, proteins and other cellular components. In order to Boolean networks, and Petri nets belong to this group. Instead,
understand complex genetic regulatory networks, a concerted continuous-state models, assume the state of each node of the net-
effort among experiments, modeling, and theory is required. In work to be a continuous function in time of the expression of the
input components. Its evolution is modeled resorting to differential
rate equations. Partial differential equations, (nonlinear) ordinary
differential equations, and linear models belong to this group. Their
main limitation consists in the difficulty of properly modeling all
夽 This work has been partially supported by Grant No. CUP B15G13000010006 biochemical reactions governing the interaction among the nodes
awarded by the Regione Valle d’Aosta for the project: “Open Health Care Network of the network.
Analysis” and by the Italian Ministry of Education, University & Research (MIUR)
Boolean networks (BN), introduced by Kauffman [10], have been
(Project PRIN 2010, MIND).
∗ Corresponding author. Tel.: +39 0110907198. proved successful in modeling real regulatory networks (e.g., see
E-mail address: [email protected] (G. Politano). [13–18]). In a BN the state of a biochemical entity, i.e., a gene, is

1877-7503/$ – see front matter © 2013 Elsevier B.V. All rights reserved.
http://dx.doi.org/10.1016/j.jocs.2013.10.005
 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344 333

described by a Boolean variable. The transition from one state to During the simulation of the network dynamics, an initialization
another state is computed by means of a Boolean function of the state sets all nodes of the network to a known configuration. The
states of other genes in the network. Transitions between states are network state is then continuously updated by repeatedly evalu-
deterministic, which means that, given an initial state, the next state ating all Boolean functions describing the network until a steady
is always the same. Although the approach seems to set a strong state or a state cycle is reached. The sequence of states traversed
simplification compared to reality, BNs enable to study high-level by a BN during the simulation of its dynamics forms a trajectory
properties of complex networks (e.g., robustness to background of the system, while the final steady state or the final state cycle
noise, behavior under different initial conditions, etc.). of the trajectory form a point or a cyclic attractor, respectively.
Recent researches suggest that several realistic biological prob- Each network is associated with a set of point and cyclic attrac-
lems may be analyzed resorting to the BN formalism. In this cases, tors, depending on its dynamics on different initialization states. In
BNs allow for a simulation of a real biological GRN’s dynamic behav- case of a point attractor, the system’s state freezes whenever the
ior (e.g., the Drosophila case), emphasizing the functional relevance network enters the attractor, and it is unable to perform further
of the resulting network topology [19]. Biologist are thus able transitions unless external perturbations are applied. Differently,
to characterize the network and evaluate its dynamics, identify- cyclic attractors show a cyclic behavior of the system. Once a trajec-
ing topological elements such as feedback circuits, i.e., cascades of tory falls into one of the states belonging the attractor, the system
regulatory interactions, negative circuits that buffer gene effects, keeps cyclically moving among the attractor’s states.
and positive circuits that may constitute developmental switches
enabling alternative developmental pathways [20,21]. Moreover, it
is possible to study properties of the biological GRN by computing
and analyzing the attractors of the network. An attractor is a single
state or a set of states towards which a system tends to converge
over time [22,23].
Most published BN based models focus on high-level gene/gene
or gene/protein interactions, neglecting post-transcriptional regu-
latory activities carried out by small non-coding RNA sequences,
such as micro RNA (miRNA).
In this paper we discuss how miRNAs and post-transcriptional
regulatory interactions can be modeled resorting to BNs. The pro-
posed BN based post-transcriptional model has been implemented
in a software tool able to simulate a BN and to compute the
attractors of the network taking into account post-transcriptional
activities [24].
We exploited the developed model and the related tool to
analyze two real networks related to Drosophila taken from [25].
Results obtained from the network analysis have been then com-
pared with experimental results discussed in [25] to show the
capability of the proposed model to provide interesting insights
related to the modeled biological process.

2. Background

Modeling general aspects of GRNs, such as genes’ interactions,

dates back to the end of 1960s. In [10], Kauffman considered an ide-
alized representation of a typical gene network. The assumption
is that genes are equivalent entities, and their interaction forms
a directed graph in which each gene receives inputs from a fixed
number of selected genes. The set of entities populating the net-
work has been recently extended including gene products (e.g.,
proteins) by [18]. The state of each regulatory entity, i.e., a gene
or a protein, is represented as a Boolean value, either 1, indicating
the presence of the entity (e.g., a gene is expressed), or 0 indicating
its absence (e.g., a gene is not expressed).
Given a set of N entities, the state of the GRN is represented
by a Boolean vector X̂ = [x1 , · · ·, xN ], leading to a state space of 2N
states. Each component xi of X̂ is described using a Boolean function
fi (xi = fi (X̂)).
To make the model computationally acceptable, the f behavior
of a BN is usually simulated in a synchronous way, i.e., all entities of
the network update their state together [13,26]:

t+1
X̂ = F̂(X̂ t ) (1) Fig. 1. Example of a simple Boolean network involving three genes (x1 , x2 and x3 )
together with the related space state. (a) The Boolean network represented as a
t+1 directed graph. Each node of the graph represents a gene while arcs represents gene
where X̂ is the next GRN state and F̂ is the vector of all functions
interactions. (b) The state space of the network represented as a directed graph
fi that map the transition of a single entity from the current state where each node represents a state of the network while arcs represent valid state
to the next one. transitions.
334 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344

To properly study the network dynamics all possible initializa- the classical RBN model. In a GPBN, gene-to-gene interactions are
tion states should be simulated. The set of initial states leading to mediated by the synthesis of proteins and other products. However
an attractor is called basin of attraction [27]. Attractor’s proper- post-transcriptional regulation carried out by miRNAs is still not
ties (length of cyclic attractors, basins dimensionality, trajectory fully considered. Fig. 2 depicts a very simple example of GRN
lengths, etc.) are commonly studied in order to infer high-level modeling a cellular regulatory activity that includes all entities
properties of a GRN [18,27]. that need to be considered in order to extend the GPBN model to
Fig. 1a gives a simple example of a BN. It contains three entities include post-transcriptional regulation mechanisms. According to
(e.g., x1 , x2 , x3 ) whose next state functions are defined as follows: the example, G1 and G2 are transcribed into two mRNA molecules
(mRNA1 and mRNA2); P1 and P2 are the resulting proteins. P2
x1t+1 = ¬x3t works as an upstream promoter of gene G3, i.e., G2 is a transcription
factor of gene G3. miRNA1 (still a product of G1) acts as a post-
x2t+1 = x1t ∧ x3t
transcriptional repressor of mRNA2, which results in a translational
x3t+1 = x2t repression of P2 and therefore in an inhibition effect on gene G3.
To properly model this post-transcriptional inter-
Fig. 1b shows the state space of this network. Each node repre- action, following the GPBN principles, we extended
sents a possible network state whereas edges represent legal state the interaction between genes (G1 and G2 of the
transitions for the network. example) by explicitly introducing their related products (as
previously done for continuous models [32,33]), thus, including
3. Models and methods miRNA1 as a product of G1.
Fig. 3 shows an extended BN representing the proposed regula-
3.1. Extending the gene protein/product Boolean network model tory example. For the sake of simplicity, the translational process
that leads to the protein production, starting from the related
Functional studies indicate that miRNAs participate in the mRNA molecule, has not been explicitly modeled. Nevertheless, if
regulation of almost every investigated cellular process like, for necessary, the model can be easily extended by adding all actors
instance, cell metabolism, signal transduction, cell differentiation, required to precisely modeling all processes. Networks nodes are
cell fates and so on [28–30]. miRNAs regulate gene expression depicted with different symbols to identify: (1) genes (circular
post-transcriptionally by interfering either with a target mRNA’s nodes), (2) mRNA Protein pairs (rectangular nodes), and (3) miRNA
translation or stability [31]. Moreover, further studies show that (rhomboidal nodes).
they can modulate mRNA–protein interactions, and suppress pro- In order to properly model the post-transcriptional regulation
tein synthesis, although the mechanistic details are still poorly mechanisms, the set of boolean functions of each transcriptional
understood. product targeted by a miRNA (e.g., the mRNA2 P2 node) must be
The interaction between mRNAs and proteins is well modeled carefully designed. Post-transcriptional regulation acts at mRNA
by the gene protein/product Boolean network (GPBN) model level, hence, considering the final protein production, it has
proposed by Graudenzi et al. in [18], which is a generalization of higher priority compared to gene expression activity. In terms

Fig. 2. Example of a GRN involving post-transcriptional mechanisms.

 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344 335

will avoid to evolve into biological illegal states. In Fig. 3 these con-
ditions are represented by the Boolean equations C1, C2, and C3.
Every time an initial state of the network is considered the three
conditions must be evaluated. A state is considered legal if all con-
ditions return zero, illegal otherwise. As an example, let us consider
G1 and the related protein mRNA1 P1. The protein can be synthe-
sized only if the related gene has been expressed. So, any state in
which mRNA1 P1 is equal to 1 (expressed), while G1 is equal to 0
(not expressed) is actually an illegal state.

3.2. Including the model into the Boolean network toolkit

We implemented the GRN model presented in Section 3.1 into a

software tool able to analyze the network dynamics by computing
Fig. 3. Boolean network model of the GRN proposed in Fig. 2 including explicit
modeling of post-transcriptional regulatory activity. the network attractors [24]. The actual implementation is based
on the Boolean network toolkit (BNT) presented in [26]. The BNT
implements the BN direct graph using adjacent lists to optimize
of Boolean functions, this can be modeled by placing the miRNA
access speed and reduce memory allocation. The core is entirely
expression state in Boolean AND with the mRNA expression
built upon the BOOST C++Library [34], which provides very efficient
state.
and cross platform libraries.
When introducing gene products such as proteins and miRNAs
Algorithm 1 shows the overall attractors search process that is
in a BN model, the synthesis and decay times for a given product
an iterative process involving probes iterations. Probes is the num-
must be taken into account. In our model, we considered an uni-
ber of initial states from which the network is simulated to identify
tary synthesis and decay time for all entities according to the GPBN
its attractors. Whenever the number N of network nodes leads to
model. Therefore, the change of state of a gene at time t influences
a state space too large to be exhaustively analyzed, it is possible to
all its products at time t + 1.
keep the computation time under control by selecting a reduced
Although the introduction of miRNA entities into the BN makes
set of randomly or user selected initial states for the network. At
it possible to account for their post-transcriptional activity into
each iteration, a candidate initial state is selected and its validity
the dynamics of the system, it is not enough to properly model
against the transcriptional and post-transcriptional constraints is
the whole post-transcriptional activity. If not properly constrained,
checked. If the state is legal, the network dynamic is simulated to
users may select initialization states for the simulation of the net-
search for an attractor on the selected trajectory.
work (that can now include combinations of genes and products)
that may be biologically not valid. To deal with illegal states, the
Algorithm 1.
description of the BN is expanded, including a set of conditions
identifying all illegal states of the network. Once a biologically valid The isValidBooleanState function (see Algorithm 2) performs
initialization state is selected, if well designed, the network itself the initial state validity check. Given the initial state to analyze

GRN attractors searching algorithm.

336 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344

Algorithm for checking the validity of an initial state.

and the set of constraints as input, this function evaluates the state 4.1. Networks description
against each constraint. If at least one of the constraints is true,
it means the given state is not valid and it needs to be discarded. The two considered GRNs that regulate the determination of
Otherwise, a true value is returned. photoreceptor cells, proprioceptor organs, and olfactory organs in
Drosophila are:
Algorithm 2.

Eventually, the tool exports all result of the network analysis 1. Photoreceptor determination network (Fig. 4a): in this network
(e.g., attractors, state space, trajectories, etc.) using the XGMML miR-7 acts in a coherent feed-forward loop cooperating with
format, which is ready to be visualized and further analyzed with Pnt-P1 in silencing YAN to create stability against fluctuations
visualization tools such as Cytoscape [35]. of Pnt-P1. This behavior is common for guaranteeing that a
cell’s fate change is not spontaneously induced or reverted [37],
4. Results and discussion admitting only oriented transitions from the state YAN ON to the
state YAN OFF in which YAN itself is fully degraded. After YAN
We tested our model against two real networks discussed in degradation, possible Pnt-P1 fluctuations will not lead to further
[25]. Both networks analyze the role of miR-7 in Drosophila. The YAN ON states.
two networks have been modeled resorting to the proposed BN 2. Sensory organ precursor (SOP) determination network (Fig. 4b): in
post-transcriptional model and then analyzed by computing the this network mir-7 belongs to an incoherent feed-forward loop.
exhaustive set of attractors and trajectories. Results obtained from This type of network motif leads to an accelerated and transient
the network analysis have been finally compared with the results pulse to downstream genes expression [38]. The overall effect is a
reported in [36] that hypothesize a stabilizing role of miR-7 against network in which fluctuating peaks of Atonal (ATO) would result
perturbations that would change the cell fate in terms of develop- in transient pulses of ATO repression by EsplC. Vice versa the
ment. This comparison provides an interesting example of the type sustained increase of ATO would result in sustained repression
of analysis and results that the proposed model is able to support. of EsplC by miR-7 and a corresponding stabilization of ATO [25].
The simulated networks described using the input for-
malism of our tool, as well as all outputs produced from the 4.2. Network modeling and attractor analysis
network analysis in standard XGMML format have been pro-
vided as additional material to this paper.1 Moreover the source Both networks presented in Fig. 4 have been partially redesigned
code of the tool used for the analysis, as well as a legacy in order to fit our extended GPBN model in which miRNAs target
Cytoscape plugin that can be used to visualize the outputs pro- proteins nodes and miRNAs are produced thanks to the expres-
duced by the tool can be freely downloaded at http://www. sion of the related hosting genes. Fig. 5 graphically shows the basic
testgroup.polito.it/index.php/bio-menu-tools/item/208-boolean- rule exploited during the BN post-transcriptional redesign pro-
regulatory-network-simulator. cess. According to [20], if the transcription/translation is active,
mRNAs/proteins are synthesized in a one time step. Thus, if the net-
work includes a miRNA node targeting a gene, a new protein node
1
We have been unable to submit the supplemental material through the EES
must be inserted in the network. The protein node must be con-
submission system. For the review process, this material is therefore available at the nected to the parent gene node and targeted by the miRNA. Also all
following URL: http://orion.polito.it/tmp/JOCS-D-13-00019-AdditionalMaterial.zip. gene products (outgoing edges) must be re-arranged accordingly.
 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344 337

their products. In order to take into account different options two

variants of each network have been considered assuming that:

• a gene is expressed when all its parent nodes are concurrently

expressed (ON): this condition models enhancer complexes com-
posed by multiple proteins that cooperate together. Resulting
Boolean functions describe this condition as a set of terms com-
puted by the ∧ operator.
• a gene is expressed when at least one of its parent nodes is
expressed (ON): this represents the behavior of multiple tran-
scription factors that autonomously induce the expression of
their targets. Resulting boolean functions describe this condition
as a set of terms computed by the ∨ operator.

In both networks, these two options have been modeled by

introducing the generic Boolean operator op ∈ { ∧ , ∨ }.
Moreover, in order to highlight the contribution of miR-7 to
the behavior of the network, we compared the two variants of the
redesigned networks reported in Figs. 6a and 7a, with two equiva-
lent networks in which the miR-7 node and related edges have been
removed (Figs. 6b and 7b).
We analyzed each network using our tool, collecting all attrac-
tors, trajectories and basins of attraction. Interestingly, the two
options introduced by the op operator leaded to different behav-
iors of the network, which still maintains the same topology. In
each simulation the exhaustive set of initial states has been ana-
lyzed with a reasonable average computation time (20–30 ms to
analyze each network).
The next sections discuss results obtained from the network
analysis. The considered networks have been simulated under sev-
eral assumptions in order to better understand the properties
of the computed attractors, along with the role of miR-7 in the
Fig. 4. Gene regulatory networks controlling photoreceptor determination and SOP network.
determination in Drosophila as published in [25]. (a) GRN controlling the photore-
ceptor determination in Drosophila as published in [25]. (b) GRN controlling the SOP
determination in Drosophila as published in [25]. 4.2.1. Photoreceptor determination network
Due to the network topology, miR-7 is expected to cooperate
with ERK and Pnt-P1 in maintaining the network steady against
YAN fluctuations.
These extensions respect the assumption that both transcription Given this premise, the comparison of the network attractors
factors and proteins undergoing post-transcriptional modifications between the network with miR-7 (Fig. 6a) and the one without miR-
decay in a one time step if their mRNAs are not present [20]. Finally, 7 (Fig. 6b), does not show any difference when the op =∧ option is
also the miRNA host gene (if not present) must be explicitly inserted selected. The set of common attractors with and without miR-7 is
in the network. reported in Table 3. In fact, the mandatory concurring presence of
While this redesign phase has been performed manually on- all silencers of YAN somehow masks the role of the miRNA. This is
going work is being carried out to make it automatic within our not surprising because, based on their discrete nature, BNs perform
simulation tool. a qualitative analysis instead of measuring the gene expression
Figs. 6a and 7a show the redesigned networks considered in rate from a quantitative perspective. This simplification makes the
our analysis. It is worth to point out here that there is no evidence analysis of complex network feasible, but it possibly masks certain
in [39] regarding the way genes cooperate in enhancing/silencing configurations and their intermediate equilibrium. In this scenario,

Fig. 5. Transformation rules of a traditional Boolean network to account for post-transcriptional activities.
338 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344

Fig. 6. Boolean network modelling the GRN responsible for Photoreceptor Determination in Drosophila presented in Fig. 4a. The network has been redesigned in order to
model post-transcriptional activity considering both the presence and the absence of miR-7. The network includes the generic Boolean operator op ∈ { ∧ , ∨ } that enables
to consider different options in the way regulation activity is performed. (a) Photoreceptor determination network with miR-7 present. (b) Photoreceptor determination
network with miR-7 absent.

the antagonist roles of Pnt-P1 and miR-7 concur in preserving the (Fig. 6b) plus a set of specific attractors. The set of common attrac-
expected pathway behavior (e.g., the complete degradation of YAN), tors is reported in Table 4. These attractors are not discussed here
possibly masking the miR-7 fine tuning effect. since they do not show any difference between the two considered
Things change when the network is analyzed under the op =∨ networks. We identified three point attractors specific of the net-
option. In this case, the tool returns 3 point attractors common in work with miR-7 and one point attractor specific of the network
both the network with miR-7 (Fig. 6a) and the one without miR-7 without miR-7:
 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344 339

Fig. 7. Boolean network modeling the GRN responsible for Sensory Organ Precursor (SOP) Determination in Drosophila presented in Fig. 4b. The network has been redesigned
in order to model post-transcriptional activity considering both the presence and the absence of miR-7. The network includes the generic Boolean operator op ∈ { ∧ , ∨ } that
enables to consider different options in the way regulation activity is performed. (a) SOP determination network with miR-7 present. (b) SOP determination network with
miR-7 absent.

• Attractors of the network with miR-7 (see Table 1): the 3 protein are suppressed, regardless the state of Notch. In fact, the
attractors confirm the expression of miR-7 and the expected suppression of EsplC is an obvious outcome because it needs the
degradation of the YAN protein. Interestingly, the three attractors simultaneous expression of both Notch and ATO (since they are
show the suppression of YAN P by the miRNA, with no regards to joined with the op =∧). The Notch silencing also leads to a sim-
the expression of the signaling genes (Notch, EGFR), and the YAN’s ilar result in the set of common cyclic attractors: all of them show
antagonists (ERK, Pnt-P1). transient pulses of ATO only when the Notch chain is turned OFF.
• Attractors of the network without miR-7 (see Table 2): the Moreover, the network with miR-7 introduces one specific point
point attractor shows no degradation of the YAN protein even attractor (see Table 5). The point attractor leads to the com-
if both YAN’s antagonists (ERK, Pnt-P1) are expressed. The ectopic plete degradation of EsplC P, while EsplC, miR-7 and ATO are still
expression of YAN P may confirm the regulatory role of miR-7 for expressed. Since only the EsplC protein is silenced whilst its encod-
the network’s stabilization. ing gene remains expressed, this attractor confirms the regulatory
role of the miRNA.
4.2.2. SOP determination network Instead, when considering the op =∨ assumption, the network
The topology of the SOP Determination Network suggests two with miR-7 (Fig. 7a) and the one without miR-7 (Fig. 7b) mani-
behaviors: (1) transient pulses of downstream genes expression, fest two common point attractors only. They show two outcomes
(2) specific polarization of upstream genes that leads to the stabi- compatible with expectancies in [39]: (i) when Notch is ON and
lization of the entire network. EGFR is OFF, miR-7 is also OFF, avoiding the silencing of EsplC P; (ii)
Let us first consider the network attractors under the op =∧ when both Notch and EGFR are ON, miR-7 is expressed, correctly
assumption. The network with miR-7 (Fig. 7a) and the one with- degrading EsplC P. The analysis of the basins of common attractors
out miR-7 (Fig. 7b) manifest a set of five point and four common provides additional information to understand these two behav-
cyclic attractors (reported in Table 6). In all of them, EsplC and its iors: the size of each basin decreases when miR-7 is introduced
340 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344

Table 1 Table 2
Unique attractors for Photoreceptor Determination Network using the operator Unique attractors for Photoreceptor Determination Network using the operator
op =∧ with miR-7. The Table shows 3-point attractors. Bold/green labels indicate op =∧ without miR-7. The Table shows 1 point attractor. Bold/Green labels indicate
ON nodes, whereas Italic/red labels indicate OFF nodes. ON nodes, whereas Italic/Red labels indicate OFF nodes.
 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344 341

Table 3 Table 4
Common attractors for Photoreceptor Determination Network using the operator Common attractors for Photoreceptor Determination Network using the operator
op =∧. The Table shows 4-point attractors. Bold/Green labels indicate ON nodes, op =∨. The Table shows 3-point attractors. Bold/Green labels indicate ON nodes,
whereas Italic/red labels indicate OFF nodes. whereas Italic/red labels indicate OFF nodes.

• Attractors of the network with miR-7 (see Table 7): the set of
attractors is composed of three point attractors. All of them
within the network, supporting the hypothesis that the presence show the expected stabilization of the network. The sustained
of the miRNA makes the network less prone to enter in one of the expression of ATO and miR-7 causes the degradation of EsplC P,
common attractors. regardless the fact that EGFR and Notch are expressed or not.
Together with the common attractors (see Table 9) the network • Attractors of the network without miR-7 (see Table 8): The three
also includes now a set of specific attractors: point attractors show the same pattern of expression: ATO, ATO P,
342 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344

Table 5
Unique attractors for SOP Determination Network using the operator op =∧ with miR-7. The Table shows 1 point attractor. Bold/Green labels indicate ON nodes, whereas
Italic/red labels indicate OFF nodes.

Table 6
Common attractors for SOP Determination Network using the operator op =∧. The Table shows 5 point attractors and 4 cyclic attractors. Bold/Green labels indicate ON
nodes, whereas Italic/red labels indicate OFF nodes.

Table 7
Unique attractors for SOP Determination Network using the operator op =∨ with miR-7. The Table shows 3 point attractors. Bold/green labels indicate ON nodes, whereas
Italic/red labels indicate OFF nodes.

Table 8
Unique attractors for SOP Determination Network using the operator op =∨ without miR-7. The table shows 3 point attractors and 4 cyclic attractors. Bold/green labels
indicate ON nodes, whereas italic/red labels indicate OFF nodes.

EsplC, and EsplC P are always turned ON, regardless of the expres- EsplC and EsplC P, unexpected in [39]. The pulses of ATO are only
sion of both EGFR and Notch. The ATO expression seems only driven by the buffer effect of Sense.
regulated by Sense because it remains still turned ON even if its
signaling chain is turned OFF. Looking at the set of four cyclic For this network only, it can be easily noticed that the op =∨
attractors, all of them share the degradation of EGFR and the option produces results that are more coherent with experimental
expression of Notch. This condition leads to the expression of both findings reported in the literature.
 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344 343

Table 9
Common attractors for SOP Determination Network using the operator op =∨. The Table shows 2 point attractors. Bold/green labels indicate ON nodes, whereas Italic/red
labels indicate OFF nodes.

Overall, given the different attractors obtained with the op =∨ • PhD wo mirna AND results.zip: Photoreceptor determination
and op =∧ operators, this preliminary work highlights how the network without miR-7 present and op =∧ option analysis and
description of the boolean function of any given node plays a crucial simulation results.
role when dealing with GRNs: a better estimate of the operator • PhD wo mirna AND.net: Photoreceptor determination network
should eventually lead to a more reliable set of attractors. without miR-7 present and op =∧ option description file.
• PhD wo mirna OR results.zip: Photoreceptor determination
network without miR-7 present and op =∨ option analysis and
5. Conclusions simulation results.
• PhD wo mirna OR.net: Photoreceptor determination network
In this paper we discussed an extended BN model to account without miR-7 present and op =∨ option description file.
for post-transcriptional regulation in GRN simulation. Thanks to • PhD wo mirna.constraints: Photoreceptor determination net-
this extended model, we discussed the set of attractors of two bio- work without miR-7 present constraint file to be used both for
logically confirmed networks, focusing on the regulatory role of simulation with op =∧ and op =∨ options.
miR-7. Attractors have been compared with networks in which the • SOP w mirna AND results.zip: SOP determination network
miRNA was removed. The central role of the miRNA for increasing with miR-7 present and op =∧ option analysis and simulation
the network stability has been highlighted in both the networks, results.
confirming the cooperative stabilizing role of miR-7. • SOP w mirna AND.net: SOP determination network with miR-7
The enhanced BN model presented in this paper is only a first present and op =∧ option description file.
step towards a more realistic analysis of the high-level functional • SOP w mirna OR results.zip: SOP determination network with
and topological characteristics of GRNs. Resorting to the tool facil- miR-7 present and op =∨ option analysis and simulation results.
ities, the dynamics of real networks can be analyzed. Thanks to • SOP w mirna OR.net: SOP determination network with miR-7
the extended model that includes post-transcriptional regulations, present and op =∨ option description file.
not only the network simulation can be more reliable, but also it • SOP w mirna.constraints: SOP determination network with
can offer new insights on the role of miRNAs from a functional miR-7 present constraint file to be used both for simulation with
perspective, and this improves the current state-of-the-art, which op =∧ and op =∨ options.
mostly focuses on high-level gene/gene or gene/protein interac- • SOP wo mirna AND results.zip: SOP determination network
tions, neglecting post-transcriptional regulations. without miR-7 present and op =∧ option analysis and simulation
Due to its discrete nature, the BN model may still neglect some results.
regulatory fine adjustments. However, the largest number of the • SOP wo mirna AND.net: SOP determination network without
computed attractors, now including miRNAs, still represents mean- miR-7 present and op =∧ option description file.
ingful states of the network. The simple glimpse into the complexity • SOP wo mirna OR results.zip: SOP determination network
of the network dynamics, that the toolkit is able to provide, could without miR-7 present and op =∨ option analysis and simulation
be used not only as a validation of in vitro experiments, but as a results.
real System Biology tool able to rise new questions and drive new • SOP wo mirna OR.net: SOP determination network without
experiments. miR-7 present and op =∨ option description file.
• SOP wo mirna.constraints: SOP determination network with-
Acknowledgement out miR-7 present constraint file to be used both for simulation
with op =∧ and op =∨ options.
The authors wish to acknowledge and thank Stefano Benedet-
tini (European Centre for Living Technology, Venice, Italy, e-mail:
[email protected]), because its knowledge and support helped References
us to develop our tool.
[1] T. Werner, Next generation sequencing in functional genomics, Brief Bioinform.
11 (5) (2010) 499–511, http://dx.doi.org/10.1093/bib/bbq018.
[2] A. Benso, S. Di Carlo, H. urRehman, G. Politano, A. Savino, Using gnome wide
Appendix A. Additional files data for protein function prediction by exploiting gene ontology relationships,
in: Automation Quality and Testing Robotics (AQTR), 2012 IEEE International
• PhD w mirna AND results.zip: Photoreceptor determination Conference on, 2012, pp. 497–502.
[3] K. Kaneko, Life: An Introduction to Complex Systems Biology, Vol. 171, Springer,
network with miR-7 present and op =∧ option analysis and sim- Heidelberg, Germany, 2006.
ulation results. [4] H.D. Jong, Modeling and simulation of genetic regulatory systems: a literature
• PhD w mirna AND.net: Photoreceptor determination network review, J. Comput. Biol. 9 (2002) 67–103.
[5] A. Csikász-Nagy, D. Battogtokh, K.C. Chen, B. Novák, J.J. Tyson, Analysis of a
with miR-7 present and op =∧ option description file.
generic model of eukaryotic cell-cycle regulation, Biophys. J. 90 (12) (2006)
• PhD w mirna OR results.zip: Photoreceptor determination net- 4361–4379, doi:10.1529/biophysj.106.081240.
work with miR-7 present and op =∨ option analysis and [6] Y. Fomekong-Nanfack, J.A. Kaandorp, J. Blom, Efficient parameter esti-
simulation results. mation for spatio-temporal models of pattern formation: case study
of drosophila melanogaster, Bioinformatics 23 (24) (2007) 3356–3363,
• PhD w mirna OR.net: Photoreceptor determination network doi:10.1093/bioinformatics/btm433.
with miR-7 present and op =∨ option description file. [7] E.P. van Someren, L.F. Wessels, M.J. Reinders, Linear modeling of genetic
• PhD w mirna.constraints: Photoreceptor determination net- networks from experimental data, Proc. Int. Conf. Intell. Syst. Mol. Biol. 8 (2000)
355–366.
work with miR-7 present constraint file to be used both for [8] N. Friedman, M. Linial, I. Nachman, Using bayesian networks to analyze expres-
simulation with op =∧ and op =∨ options. sion data, J. Comput. Biol. 7 (2000) 601–620.
344 G. Politano et al. / Journal of Computational Science 5 (2014) 332–344

[9] E.J. Moler, D.C. Radisky, I.S. Mian, Integrating naive bayes models and external [38] S. Mangan, A. Zaslaver, U. Alon, The coherent feedforward loop serves as a sign-
knowledge to examining copper and iron homeostasis in s. cerevisiae, Physiol. sensitive delay element in transcription networks, J. Mol. Biol. 334 (2) (2003)
Genom. 4 (2) (2000) 127–135. 197–204.
[10] S.A. Kauffman, Metabolic stability and epigenesis in randomly constructed [39] X. Li, J.J. Cassidy, C.A. Reinke, S. Fischboeck, R.W. Carthew, A microrna imparts
genetic nets, J. Theor. Biol. 22 (3) (1969) 437–467. robustness against environmental fluctuation during development, Cell 137 (2)
[11] R. Albert, Boolean modelingof genetic regulatory networks, in: E. Ben-Naim, H. (2009) 273–282, http://dx.doi.org/10.1016/j.cell.2009.01.058.
Frauenfelder, Z. Toroczkai (Eds.), Complex Networks, Vol. 650 of Lecture Notes
in Physics, Springer, Berlin Heidelberg, 2004, pp. 459–481.
[12] L.J. Steggles, R. Banks, A. Wipat, Modelling and analysing genetic networks: Gianfranco Politano received the MS degree in computer
from Boolean networks to petri nets, in: Proceedings of the 2006 International engineering and the PhD degree in information technolo-
Conference on Computational Methods in Systems Biology, CMSB’06, Springer- gies from the Politecnico di Torino, Italy, where he has
Verlag, Berlin, Heidelberg, 2006, pp. 127–141. been a postdoc in the Department of Control and Com-
[13] J. Bower, H. Bolouri, Computational Molecular Biology Series, MIT Press, 2001 puter Engineering since 2011. His main research topics are
http://books.google.it/books?id=-9hZDDfMroQC system reliability and machine learning techniques. He is
[14] P. Rämö, J. Kesseli, O. Yli-Harja, Perturbation avalanches and criticality in a full member of the IEEE and the IEEE Computer Society.
gene regulatory networks, J. Theor. Biol. 242 (1) (2006) 164–170, http://dx.doi.
org/10.1016/j.jtbi.2006.02.011.
[15] R. Serra, M. Villani, A. Graudenzi, S.A. Kauffman, Why a simple model
of genetic regulatory networks describes the distribution of avalanches
in gene expression data, J. Theor. Biol. 246 (3) (2007) 449–460,
http://dx.doi.org/10.1016/j.jtbi.2007.01.012.
[16] S. Bornholdt, Boolean network models of cellular regulation: Alessandro Savino received the MS degree in computer
prospects and limitations, J. R. Soc. Interf. 5 (Suppl. 1) (2008) S85–94, engineering and the PhD degree in information technolo-
doi:10.1098/rsif.2008.0132.focus. gies from the Politecnico di Torino, Italy, where he has
[17] B. Wilczynski, E.E.M. Furlong, Challenges for modeling global gene regulatory been a postdoc in the Department of Control and Com-
networks during development: insights from drosophila, Dev. Biol. 340 (2) puter Engineering since 2009. He is also associated with
(2010) 161–169, http://dx.doi.org/10.1016/j.ydbio.2009.10.032. IRCC - Institute for Cancer Research and Treatment of Can-
[18] A. Graudenzi, R. Serra, M. Villani, C. Damiani, A. Colacci, S.A. Kauff- diolo (TO). His main research topics are microprocessor
man, Dynamical properties of a boolean model of gene regulatory test and software-based self-test as well as bioinformatics
network with memory, J. Comput. Biol. 18 (10) (2011) 1291–1303, and image processing.
http://dx.doi.org/10.1089/cmb.2010.0069.
[19] R. Thomas, R. D’Ari, Biological Feedback, CRC Press, Boca Raton, FL, 1990.
[20] R. Albert, H.G. Othmer, The topology of the regulatory interactions predicts the
expression pattern of the segment polarity genes in Drosophila Melanogaster,
J. Theor. Biol. 223 (2003) 1–18. Alfredo Benso received the MS degree in computer engi-
[21] L. Sánchez, D. Thieffry, A logical analysis of the drosophila gap-gene system, J. neering and the PhD degree in information technologies,
Theor. Biol. 211 (2) (2001) 115–141, http://dx.doi.org/10.1006/jtbi.2001.2335. both from Politecnico di Torino, Italy, where he is working
[22] S. Huang, I. Ernberg, S. Kauffman, Cancer attractors: a systems view of as a tenured associate professor of computer engineering.
tumors from a gene network dynamics and developmental perspective, His current research interests include bioinformatics, and
Semin. Cell Develop. Biol. 20 (7) (2009) 869–876, http://dx.doi.org/10. system biology. He is also actively involved in the Com-
1016/j.semcdb.2009.07.003. puter Society, where he has been a leading volunteer for
[23] J.X. Luo, M.S. Turner, Evolving sensitivity balances Boolean networks, PLoS One several projects. He is a Computer Society Golden Core
7 (5) (2012) e36010, http://dx.doi.org/10.1371/journal.pone.0036010. Member, and a senior member of the IEEE.
[24] A. Benso, S. Di Carlo, H. urRehman, G. Politano, A. Savino, Accounting for post-
transcriptional regulation in Boolean networks based regulatory models, in:
IWBBIO 2013, International Work-Conference on Bioinformatics and Biomed-
ical Engineering, 2013, pp. 397–404.
[25] S. Artavanis-Tsakonas, M.D. Rand, R.J. Lake, Notch signaling: cell fate control Stefano Di Carlo received the MS degree in computer
and signal integration in development, Science 284 (5415) (1999) 770–776. engineering and the PhD degree in information technolo-
[26] S. Benedettini, M. Villani, A. Roli, R. Serra, M. Manfroni, A. Gagliardi, gies from the Politecnico di Torino, Italy, where he has
C. Pinciroli, M. Birattari, Dynamical regimes and learning properties of been an assistant professor in the Department of Con-
evolved Boolean networks, Neurocomputing 99 (0) (2013) 111–123, trol and Computer Engineering since 2008. His research
http://dx.doi.org/10.1016/j.neucom.2012.05.023 http://www.sciencedirect. interests include DFT, BIST, and dependability as well as
com/science/article/pii/S0925231212004870 bioinformatics and system biology. He is a golden core
[27] M. Aldana, S. Coppersmith, L. Kadanoff, Boolean dynamics with random member of the IEEE Computer Society and a member of
couplings, in: E. Kaplan, J. Marsden, K. Sreenivasan (Eds.), Perspectives the IEEE.
and Problems in Nolinear Science, Springer, New York, 2003, pp. 23–89,
http://dx.doi.org/10.1007/978-0-387-21789-5 2.
[28] D.G. Hendrickson, D.J. Hogan, H.L. McCullough, J.W. Myers, D. Her-
schlag, J.E. Ferrell, P.O. Brown, PLoS Biol 7 (11) (2009) e1000238,
http://dx.doi.org/10.1371/journal.pbio.1000238.
Hafeez Ur Rehman received the MS degree in computer
[29] N. Cloonan, M.K. Brown, A.L. Steptoe, S. Wani, W.L. Chan, A.R.R. Forrest, G.
engineering from Politecnico di Torino, Italy. He has been
Kolle, B. Gabrielli, S.M. Grimmond, The mir-17-5p microrna is a key regula-
a PhD student in the Department of Control and Com-
tor of the g1/s phase cell cycle transition., Genome Biol 9 (8) (2008) R127,
puter Engineering since 2011. His main research topic is
http://dx.doi.org/10.1186/gb-2008-9-8-r127.
proteomic and in particular proteins function prediction.
[30] A. Benso, S. Di Carlo, G. Politano, A. Savino, A new mirna motif protects path-
ways’ expression in gene regulatory networks, in: IWBBIO 2013, International
Work-Conference on Bioinformatics and Biomedical Engineering, 2013, pp.
377–384.
[31] H. Hwang, T. J. H. University, Dynamic Regulation of MicroRNAs by
Post-transcriptional Mechanisms, Johns Hopkins University, 2009 http://
books.google.it/books?id=BWvTZ6 NU0YC
[32] L. Glass, S.A. Kauffman, Co-operative components, spatial localization and oscil-
latory cellular dynamics, J. Theor. Biol. 34 (2) (1972) 219–237.
Alessandro Vasciaveo received the MS degree in com-
[33] L. Glass, S.A. Kauffman, The logical analysis of continuous, non-linear
puter engineering form Politecnico di Torino in 2012. His
biochemical control networks, J. Theor.l Biol. 39 (1973) 103–129,
main research interests are evolutionary algorithms and
http://dx.doi.org/10.1016/0022-5193(73)90208-7.
biological data modeling.
[34] B. Dawes, D. Abrahams, R. Rivera, Boost C++libraries, [Available online]:
http://www.boost.org/doc/libs/
[35] Cytoscape, Cytoscape An open source platform for complex network analysis
and visualization, [Available Online]: http://www.cytoscape.org
[36] E. Hornstein, N. Shomron, Canalization of development by microRNAs, Nat.
Genet. 38 (Suppl) (2006) S20–4, doi:10.1038/ng1803.
[37] G.J. Melen, S. Levy, N. Barkai, B.-Z. Shilo, Threshold responses to mor-
phogen gradients by zero-order ultrasensitivity, Mol. Syst. Biol. 1 (2005),
http://dx.doi.org/10.1038/msb4100036, 2005.0028.