Sei sulla pagina 1di 10

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

Application of Stochastic Simulations of Biological Networks Under Different


Impulsive Scenarios
Mge Yazc1, Vilda Purutuolu2, Derya Altntan3
1,2
Department of Statistics, Middle East Technical Universtiy, Turkey
3
Department of Mathematics, Seluk University, Turkey
mugergul15@gmail.com

ABSTRACT
The impulses are one of the sources of fluctuations
which lead to the sharp changes in the biochemical
systems such as the changes in epidemic and
population models. These effects have been
comprehensively studied under deterministic
models, whereas, have not been analyzed in
stochastic models. Hereby, in this study, we
implement these effects in two exact stochastic
simulation algorithms, namely, the direct method
and the first reaction method under different
dimensional systems and impulses scenarios. In the
application, we use real systems and apply impulses
to the biologically meaningful proteins. Then, we
evaluate the performances of both exact methods in
terms of computational demand via distinct criteria.

KEYWORDS
Stochastic simulation, biochemical systems,
impulses, direct method, first reaction method.

1 INTRODUCTION
The biochemical modeling is the mathematical
way to explain the dynamics of biological
systems such as the interaction of proteins and
molecules (Wilkinson, 2006). But, the
vivo/vitro analyses of the biological systems
can be expensive and limited. In this situation,
we can perform the simulation methods under
realistic assumptions so that we can visualize
the behavior of the real biological system
artificially.
Each biological event can be represented by a
reaction. Thus, the biochemical reaction is a

ISBN: 978-1-941968-35-2 2016 SDIWC

way to express a biological activation so that it


can be presented by a mathematical expression,
as shown below.
(1)
where ma and mb are called the stoichiometric
coefficient of the reactant A and the product B,
respectively. Furthermore, k is the rate constant
of the reaction which is dependent on the
temperature and the volume of the
environment.
Hereby, in this study, we aim to compare the
performance of the stochastic simulation
algorithms of the biological systems and to
evaluate their performances under abrupt
changes. In real systems, such abrupt changes
are called as impulses and in biological
networks, these changes are used to present the
cancer, diabetes or similar diseases whose
attacks
are
observed
suddenly
and
stochastically (Jonge, 2014).
Therefore, in the application, we choose distinct
dimensional real systems. In the calculation, we
apply the Lotka Volterra model as the toy
example. Then, we simulate the PKC pathway
and LysisLysogeny model as moderate networks.
Finally, we use the sole JAK-STAT pathway
and the combination of JAK-STAT and PKC
pathways as a single and large system. In the
simulation, we perform the direct and the first
reaction algorithms which are the two wellknown exact stochastic simulation algorithms.
In both methods, the assumption is that the

82

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

molecules of the species, which constitute a


system, mix well in a constant volume and
under a constant temperature. Accordingly, in
the organization of the study, we introduce the
underlying simulation algorithms and the
impulses in Section 2. Then, we present the
measurement criteria and the findings of the
implementations in Section 3. Finally, we
conclude the study and discuss the future work
in Section 4.

computationally efficient in simulations of


large networks as the calculation takes long
time.
The Gillespie algorithm is known as the
discrete event simulation. For a system with r
reactions and n species, its procedure can be
listed as follows.
i.

The system at time t=0 with the rate


constants c1, c2,.,cr and the molecular
numbers for each species x1, x2,., xn
are initialized.

ii.

The hazard functions, hi(x,ci) are


computed for each reaction (i=1,,r)
and the total hazard is calculated via
h0(x,c)=
.

iii.

The time interval for the next event, i.e.,


reaction, is generated from the
exponential density with a rate h0, i.e.,
Exp(h0(x,c)).

iv.

The reaction is simulated from the


mathematical
distribution
with
probabilities hi(x,ci)/h0(x,c) under the
assumption that the reactions are
independent of each other.

v.

The outputs x and t are updated via


x:=x+S(j) and t:=t+t, respectively. Here
S(j)(j=1,,n) indicates the jth column of
the stoichiometry matrix and t denotes
the new time interval used in the update
of the current time t. In the
stoichiometry matrix, the rows show the
reaction and the columns denote the
number of species. This matrix has rn
dimension for a system with r reactions
and n species.

vi.

If the updated time is less than the total


time of the simulation T, i.e., t<T, then,
the algorithm is repeated from step (ii)
until t T.

2 METHODS
2.1 Stochastic Simulation Algorithms
The gene regulation is generally modeled by
differential equations (DE). Although this
deterministic model is the most widely used
approach and can successfully explain the
steady-state behavior of the system under
discrete time and states, it cannot describe the
randomness of the system as seen in the
transcription and the translation of the proteins.
Thus, the stochastic model is the only choice to
present this randomness in a mathematical way
(Bower and Bolouri, 2001; Purutuolu, 2010).
There are five main exact stochastic simulation
algorithms which enable to generate the actual
randomness in the model. These are the direct
method (Wilkinson, 2006), also called Gillespie
method, the first reaction method (Gillespie
1992; Turner, 2005), the next reaction method
(Gibson and Bruck, 2000; Turner, 2004), also
named as the Gibson-Bruck algorithm, the
optimal direct method (Cao, 2004) and the
sorting direct method (McCollum and Peterson,
2006).
2.1.1 Direct Method
The direct algorithm, also called as the
Gillespie method, is the fastest simulator in the
calculation of time, especially, for small
systems (Wilkinson, 2006, Gillespie, 1977;
Turner, 2004). However, its application is hard
in heterogeneous situation and is not

ISBN: 978-1-941968-35-2 2016 SDIWC

83

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

2.1.2 First Reaction Method


The first reaction method is a simulation
method which is suggested to decrease the
computational time of the Gillespie algorithm
for large systems (Gillespie, 1992; Turner,
2004).
The steps of this algorithm for a system with r
reactions and n species are presented as below.
i.

The rate constants c1,..,cr, the starting


numbers of molecules for species
x1,,xn and the total time of the
simulation T are initialized for the
system at time t=0.

ii.

The hazard function for each reaction hi


(i=1,,r) is calculated.

iii.

The time interval for each reaction ti


(i=1,,r) is computed from the
exponential distribution with a rate
hi(x,ci), i.e, ti~Exp(hi(x,ci)).

iv.

The smallest ti is chosen as a time


interval for the next reaction which is
generated from the multinominal
distribution with probabilities hi/h0 as
taken in the Gillespie algorithm,
h0=
.

v.

All numbers of molecules x and the time


t are updated via t:=t+tj and x:=x+S(j) in
which S(j) denotes the jth row of the
stoichiometry matrix.

vi.

If t<T, step (ii) is returned and the


algorithm is repeated until t T.

As seen from the procedure that both the


Gillespie and the first reaction algorithms are
similar to each other. But the latter does not use
the total hazard in the update of the next
reaction time. Hence, it can produce longer

ISBN: 978-1-941968-35-2 2016 SDIWC

time steps, resulting in less computation


demand than the Gillespie method.
2.2 Impulses
The natural activity is generally dynamic and
can have sudden changes which are called as
the impulses. The shocks, natural disasters and
vaccinations are some examples of the
impulses. Hereby, these dynamic changes can
happen into two different ways, namely,
deterministically and stochastically. If the
concentration of species in the system is low
and the stochastic fluctuations have a
significant effect on the system, the stochastic
impulses are used to describe the sudden
changes in the model.
The deterministic impulses are simply based on
the ordinary differential equations (ODE) and
the random behavior is ignored in this
approach. In order to define the time of
impulses, the hybrid strategies are used. The
hybrid methods basically split the species and
the reactions as continuous and discrete groups.
The former describes the fast reactions and
associated species and the latter represents
relatively slow reactions and related species.
On the other hand, the time of impulses in the
deterministic approach is called as the jump
times that are found by the discrete groups of
reactions in the implementation of the hybrid
methods (Bower and Bolouri, 2001).
The simple ODE system can be summarized as
shown in Equation 2.
dx (t )
f ( x(t ))
dt

(2)

where x(t ) ( x1 (t ), x2 (t ),, xn (t )) is the state


vector at time t>0 when the system is defined
by the n number of species and the r number of
reactions as used beforehand. Accordingly, the
deterministic impulsive model is based on this
general ODE modeling with a jump function,

84

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

called impulse function, as indicated in


Equations 3 and 4 for a given initialization of
the system via the (t0, x0) (x0=x01,.,x0n). Here,
t0 and x0 refer to the initial time and the
associated initial state, respectively.
(3)
(4)
In Equation 4,
denotes the impulsive
function at time
. When the impulse
occurs at time
, the state vector x(t) =
x(t;t0,x0) is transferred to the new position x(t)
= x(t;
) where
is the new x(t)
point for
).
On the other side, the stochastic impulses are
based on the chemical master equations with
impulses at fixed or varying times. Hereby, the
time of impulses can be generated under
different scenarios. For instance, it can be
produced as a continuous function, threshold
function or a fixed value from a Poisson
distribution. For the continuous function, the
impulse can occur at time t=
, where x is
the r-dimensional vector of x. For the threshold
function, the time of jumps depends on a
threshold defined by xT. Here, x(t) and xT
denote the current time and the state of the total
time T where t: x(t) xT. Finally, in the last
case, the time of jumps is determined from a
random value from a Poisson distribution
(Altntan et al., 2010). In this study, we merely
deal with the first two scenarios.

Average search depth: A biochemical system


has many reactions such as j = 1,2,r. So, the
average search depth, SD, indicates the average
number of operations in the firing of a reaction
as shown below.
SD=

(5)

In Equation 2, j represents which reaction


occurs during the simulation and j stands for
the time of the occurrence of the jth reaction.
Average weighted degree: The average
weighted degree, WD, calculates the average
degree in the firing of each reaction. Here, the
degree implies how many reactions are affected
by the firing of a reaction. Thereby, WD is
computed as follows.
WD=

(6)

indicates the degree of the jth node and j


describes the number of firing times of the jth
reaction as presented in the expression of SD.
Central processing unit: The central
processing unit, CPU, is the necessary unit to
understand and to translate the program codes.
Therefore, it is also known as the computer
time of a calculation.
3.2 Description of Systems

3 APPLICATION
3.1 Measurement Criteria
In order to assess the performance of different
simulation methods in terms of the
computational demand, certain measures are
suggested. Among alternatives, the most widely
used ones can be listed as the average search
depth, the average weighted degree and the
central processing unit.

ISBN: 978-1-941968-35-2 2016 SDIWC

In order to evaluate the performance of


stochastic simulation algorithms, we select
different dimensional systems, namely, small,
moderately large and realistically large
networks. For the toy system, we choose the
Lotka-Volterra model (Wilkinson, 2006). This
system is based on the relation of preys and
predators. On the other hand, for the
moderately large system, we select the reaction
list of the PKC pathway. This system stands for

85

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

a model of the protein kinase C signal


transduction pathway, related to the significant
neural functions, especially, for the memory
and the learning process. Moreover, it has
importance at neuronal functions such as the
synaptic long-term potentiation (LTP) and the
depression (LTD) (Manninen et al., 2006). On
the other hand, as the moderate system, we also
use the Lysis-Lysogency (LL) model (Arkin et
al., 1998). Indeed, the lysogenic cycle implies
the reproduction of viral issues and the lytic
means that the cell integrity is distorted. The
mechanism of the lytic system is initiated with
the transcription and the translation. Generally,
this system has low concentrations of species
and slow reaction rates. Furthermore, we also
apply the JAK-STAT pathway (Maiwald et al.,
2010) under the realistic complexity. This
network is one of the major signaling pathways
controlling the immune system based on the
regulation of interferons. The interferons
(IFNs) are the proteins synthesized by host
cells in the reply to pathogens like viruses or
bacteria and are used for the bridge between
cells to activate defenses of the immune system
against pathogens. Finally, we investigate the
convergence of JAK-STAT and PKC pathways
to observe the response of the methods for large
systems. In the activation, the reaction of the
STAT1 and GATA4 proteins trigger the
underlying convergence. The vasoactive
hormone, called AII, changes the transcription
of the gene. AII reacts with AT1R and AT1R
modifies the biological effects of AII. AT1R is
generally used for the treatment of the
cardiovascular disease. In fact, some STAT
molecules are activated by AII. Hereby, the link
of these two pathways are based on AT1R and
AII activations (Wang, 2005).
3.3 Results
The LV, LL, PKC and JAK-STAT systems are
investigated for the time t= 5, 20 and 50 for
both the direct and the first reaction methods. In
simulations, the initial number of molecules is
taken as 100 arbitrarily for each system.

ISBN: 978-1-941968-35-2 2016 SDIWC

Because there is no stable biological knowledge


about the initial number of molecules of each
gene. Whereas, the reaction rates of each
system are set to their biologically validated
values in the associated literature about every
system. The list of reactions and the proteins in
the system are described in details in the study
of Wilkinson (2006) for the LV pathway, in the
study of Schaechter and Benowitz (1993) for
the PKC pathway in the study of Arkin et al.,
(1998) for the LL pathway and in the study of
Maiwald et al., (2010) for the JAK-STAT
pathway.

Figure 1: The plots indicate the changes in the number


of the PKC-cytosolic, Substance PKC.Ca2+ and
PKC.DAG.AA of the PKC system via Direct Method
(DR) and First Reaction Method (FR) for the time 20
when the system has no impulse.

In Figure 1, we present the changes in the


number of substances for the PKC system with
t=20 as illustration. From the results, it is
observed that the outputs of both the direct and
the first reaction methods are almost the same
apart from the fluctuations due to the
stochasticity of the systems.
Then, we investigate the impulsive effects in all
the systems by using these two methods for
short and long runs. In order to observe
underlying effects, we consider two different

86

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

scenarios for each system. The first scenario is


to detect the effect of the changes in the number
of molecules at a decreasing direction. Hereby,
if the number of the biologically meaningful
substance becomes greater than 60 in
simulations, we decrease it 20 units. On the
other side, in Scenario 2, we consider to
decrease the number of this substance at 20
units if it exceeds 60 molecules in the
simulations. For the PKC and JAK-STAT
systems, the impulses are put on STAT2c and
ISGF-3c_CP substances, respectively, as they
control whole activations in these pathways.

Figure 2: (Scenario 1) The plots indicate the changes in


the number of the Receptor IFNAR1, STAT2c and ISGF3c_CP of JAK-STAT system via Direct Method (DR)
and First Reaction Method (FR) for the fixed time 20
when the system has the impulsive effect.

ISBN: 978-1-941968-35-2 2016 SDIWC

Figure 3: (Scenario 2) The plots indicate the changes in


the number of Receptor IFNAR1, STAT2c and ISGF3c_CP of JAK-STAT Model for Direct Method (DR) and
First Reaction Method (FR) for the fixed time 20,
respectively, when the system has the impulsive effect.

In Figures 2-3, we show the plots of the


changes in activations for Receptor IFNAR1,
STAT2c and ISGF-3c_CP proteins in the JAKSTAT pathways under impulses as examples.
From the plots, we see the considerable effects
of impulses in the selected proteins. On the
other hand, similar to previous findings, there is
no biological differences in both simulation
algorithms. On the other side, when we
compare the tabulated values for all systems as
given in Table 1, we observe that the impulses
cause sharp increases in the computational
demand for small systems in terms of the local
time and CPU. Furthermore, the impulses under
Scenario 1 (increase in the number of
molecules) have less WD and higher SD than
impulses under Scenario 2 (decrease in the
number of molecules) and the values of WD
and SD under no impulses are roughly equal to
the average of the systems scores of both
impulses. Finally, there is no significant
difference between the results of both
algorithms for all time periods as seen in Table
1.
On the other hand, if we compare the findings
of the moderate systems as presented in Tables
2-4, we detect that there is no unique case
where the results of impulses produce higher
scores within the impulsive scenarios and
between the scenario with/without impulses.
For instance, the values of SD and WD are
higher under the impulses of Scenario 1
regarding the corresponding entries of Scenario
2 and both SD and WD are smallest when there
is no impulse for the PKC pathway as given in
Table 3. On the contrary, SD without impulses
is the highest compared to both sorts of

87

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

impulsive outputs and there is no tendency in


the entries of SD and WD within the scores of
impulsive scenarios for the JAK-STAT
pathway as shown in Table 4. Moreover, we
observe that the outputs of the direct method
are better in all cases for the PKC pathway,
whereas, it is not seen any particular improved
performance of the direct method for the JAKSTAT pathway.
Finally, from the results of the large system as
presented in Table 5, both SD and WD are
higher when there is no impulse in the system.
Furthermore, CPU and local time increase
significantly under impulses as found in all
previous findings. But the main gain of the first
reaction method over the direct one in the
computational demand can be better observed
when the system becomes larger and the
duration of the simulation is taken longer. In
this study, all the codes for both algorithms and
their impulsive versions are originally
developed in R by the authors and we do not
use any particular CRAN package developed
for the simulations such as smfsb or
GillespieSSA. Hence, we observe that the direct
method typically has less CPU and uses less
local time in the calculation with respect to the
results of the first reaction method for small and
moderate systems with/without impulses, and
this conclusion is still valid if the system is
large, but the time of the simulation is longer.
On the other side, the first reaction method over
performs when the system is complex and the
duration of the simulation is longer. This result
is invariant from the firing of the impulses as
well as their types, i.e., scenarios. Lastly, if we
compare the outputs of without impulses versus
with impulses, it is observed that the values of
CPU and local time are always bigger under the
impulsive effects for all dimensions and
computational time. Whereas, SD and WD
values are affected by the types of impulses and
there is no unique conclusion in these criteria
with respect to the findings of without impulses
under distinct dimensions and large systems, as
well as different simulation times.

ISBN: 978-1-941968-35-2 2016 SDIWC

4 CONCLUSION
In this study, we have dealt with the stochastic
simulation of the biochemical systems by
performing two major algorithms in this field.
These are the direct and the first reaction
methods. They are exact algorithms in the sense
that they are based on the chemical master
equation. Here, we have evaluated them under
different dimensional systems and impulsive
scenarios.
From the results, we have observed that there is
no significant difference between the direct and
the first reaction methods in most of the
pathways and simulation times. But a slight
difference is seen when the system gets
complicated. Moreover, we have detected that
the effect of impulses can be observable for
large systems and can be hardly seen under
small or moderate dimensional systems.
Furthermore, if the impulses are put on the
initial species which can trigger whole
activation of the systems, such as the growth
factor or external stimulus, the activations of all
remaining species alter too as expected and the
impulses become visible in the outputs of the
simulations. On the other hand, if the
underlying effects, either in increasing or
decreasing direction in the number of
molecules, are frequently fired, they cannot be
observed clearly. Whereas, such effects have
always caused an increase in CPU and local
time as expected and these differences become
visible while the systems have complex
structures, i.e., higher dimensions.
As a future work, we consider to investigate the
effect of bifurcations in the stochastic modeling
and in these algorithms. Because the impulses
can cause unstable positions for the systems
and the bifurcation is detected under such
instabilities. Finally, we think to extend this
study for the approximate stochastic simulation
algorithms under various conditions of

88

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

2.68

2.69

2.72

2.64

2.66

2.75

First Reaction
Method
Without Impulse

2.07
13511.13
1485.81

First
Reaction
Method

Local Time

SD

WD

t=5

16.20

352.26

5.49

2.78

t=20

6677.65

29867.95

5.68

2.99

t=50

8806532.30

2935560.26

457.51

231.66

t=5

22.88

388.73

5.64

2.78

t=20

11153.05

45008.64

6.70

2.99

t=50

9899785.86

6675700.8

546.5

242.8

2.76
13517.28
1486.68
2.33
2.63
4132.7
30.52
t=50

ISBN: 978-1-941968-35-2 2016 SDIWC

Direct
Method

CPU

2.20

2.04
11798.44
1441.96
26.42
t=20

3922.02

2.58

2.36

1483.75

13278.22

2.73

2.22

1.95
12095.05
1445.47
22.24
t=5

3651.37

2.38

2.57

1475.25

12961.56

2.55

2.39

2.06
12280.33
1447.36
28.89
t=50

4127.22

2.62

2.33

1485.81

13511.13

2.79

2.20

2.04
11789.94
1440.7
24.69
t=20

3916.6

2.59

2.35

1482.68

13274.68

2.72

2.23

1.97
12089.78
1444.55
20.98
t=5

3648.11

2.37

2.55

1474.64

12959.4

2.59

2.39

SD
Local Time
CPU
SD
Local
Time
CPU

LV
Path
way

CPU

Local
Time

SD

WD

With Impulse (scenario 1)


Without Impulse

Table 2: The results of Lysis-Lysogency (LL) pathway


for the Direct and the First Reaction methods.
LL Pathway

WD

With Impulse (scenario 2)

WD

Table 1: The results of the Lotka-Volterra (LV)


pathway for the Direct and the First Reaction Methods

Direct Method

impulses and investigate the effect of impulses


in the leap condition.

89

ISBN: 978-1-941968-35-2 2016 SDIWC


170.35
478.39
1221

t=5

First Reaction
t=20
Method

t=50

24.22
22.34
23.01
23.76

8329.5
9107.82

t=5 1286.39

First Reaction
t=20 1444.1
Method

t=50 1776.67 10587.56

22.06

SD

t=50 1608.52 9945.06

8245.74

Local
Time

23.24

1259.2

9.28

9.53

10.24

9.29

9.39

Without Impulse

6299.52

3026.31

1471.29

4730.27

2526.77

10.21

SD

t=20 1368.56 8870.04

Direct
Method

t=5

851.9

t=50

CPU

323.88

Direct Method t=20

1348.92

Local
Time

Without Impulse

7.11

7.17

7.18

7.03

7.18

7.18

WD

5.70

5.75

5.91

5.72

5.73

5.87

WD

184537.8

185542.4

187302.8

184121.8

185175.6

187167.2

Local Time

10.46

10.50

10.97

10.47

10.47

10.85

SD

Local
Time
SD

2262.23

2028.8

1884.21

21.50

97368.3

23.27

96482.65 22.48

95327.5

2135.85 96976.58 23.38

1959.43 96219.07 22.43

1862.15 95208.14 21.93

CPU

WD

5.85

5.86

5.96

5.85

5.87

5.94

7.00

7.12

7.15

6.99

7.17

7.11

WD

With Impulse (Scenario 1)

3282.69

3511.34

3848.27

3151.68

3405.7

3807.95

CPU

With Impulse (Scenario 1)


Local
Time

190665

188761
9.19

9.38

9.94

SD

5.75

5.80

5.87

WD

9.75

9.98

5.76

5.81

Local
Time

SD

2777.31 182215.3 22.56

2458.54 98606.32 22.92

2316.78 97707.35 22.07

2600.19 99525.47 22.78

2386.79 98328.78 22.95

2294.29 97610.94 21.61

CPU

7.25

7.18

7.18

7.28

7.22

7.18

WD

With Impulse (Scenario 2)

4877.44 191787.9

4201.15 189185.1

3934.43 187927.7 10.54 5.88

4536.39

4073.76

3894.72 187801.2

CPU

With Impulse (Scenario 2)

Table 3: The results of the PKC pathway for the Direct


and the First Reaction methods.

JAK-STAT
Pathway

133.55

CPU

t=5

PKC Pathway

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

Table 4: The results of the JAK-STAT pathway for the


Direct and the First Reaction methods.

90

Proceedings of The Third International Conference on Data Mining, Internet Computing, and Big Data, Konya, Turkey 2016

Table 5: The results of the convergence of JAK-STAT


and PKC systems for the Direct and the First Reaction
methods.

[7]

Jonge, Peter, Alonso, Jordi, Stein, Dan, Kiejna,


Andrzej, Aguilar-Gaxiola, Sergio, Viana and Maria,
Carmen et al. (2014). Associations between DSM-IV
mental disorders and diabetes mellitus: a role for
impulse control disorders and depression,
Diabetologia, 57 (4), 699709.

[8]

Jun, Wang (2005). Convergence of Protein Kinase C


and JAK-STAT Signaling on Transcription Factor
GATA-4. Molecular and Cellular Biology, 25 (22),
98299844.

[9]

Maiwald, Tim, Schneider, Annette, Busch, Hauke,


Sahle, Sven, Gretz, Norbert, Weiss, Thomas
S.,Kummer, Ursula, and Klingmller, Ursula.
(2010). Combining theoretical analysis and
experimental data generation reveals rf9 as a crucial
factor for accelerating nterferon alpha-nduced early
anti viral signalling. The FEBS Journal, 277 (22),
474154.

[10]

Manninen, Tiina and Makiraatikka, Eeva (2006).


Discrete stochastic simulation of cell signaling:
comparison of computational tools. Conference
Proceedings IEEE Engineering Medicine and
Biology Society, 1, 2013-2016.

[11]

McCollum, James, M. and Peterson, Gregory, D.,


Cox, Cris, Simpson, Michael, Samatova, Nagiza
(2006). The sorting direct method for stochastic
simulation of biochemical systems with varying
reaction execution behavior. Computational Biology
and Chemistry, 30 (1), 39-49.

[12]

Purutuolu, Vilda (2010). Survey on stochastic


simulations for biochemical reactions (Biyokimyasal
reaksiyonlar
iin
stokastik
simulasyon
algoritmalarna genel bir bak). Journal of Statistical
Research, 7, 70-82.

[13]

Purutuolu, Vilda and Ayyldz, Ezgi (2014).


Statistics for Bioinformatics (Biyoinformatik
Alannda statistik). Nobel Press. Ankara.

[14]

Turner, Thomas, Schnell, Santiago and Burrage,


Kevin. (2004). Stochastic approaches for modelling
in vivo reactions. Computational Biology and
Chemistry, 28, 165178.

[15]

Wilkinson, Darren J. (2011). Stochastic Modelling


for Systems Biology, Chapman and Hall Press/ CRC
Press.

ACKNOWLEDGEMENT
The authors would like to thank the AGEP
project of the Middle East Technical University
(Project no: BAP-08-11-2014-007) for their
support.
REFERENCES
[1]

[2]

[3]

Altntan, Derya, Purutuolu, Vilda and Uur, mr


(2014). Impulsive expression in stochastic
simulation algorithms (Under review).
Arkin, Adam, Ross, John and McAdams, Harley
(1998). Stochastic kinetic analysis of developmental
pathway bifurcation in phage -nfected escherichia
coli cells. Genetic, 149 (4), 1633-1648.
Bower, James M. and Bolouri, Hamid (2001).
Computational
Modeling
of
Genetic
and
Biochemical Networks, MIT Press.

[4]

Gibson, Michael A. and Bruck, Jehoshua (2000).


Efficient exact stochastic simulation of chemical
systems with many species and many channels.
Journal of Physical Chemistry, 104 (9), 1876-1889.

[5]

Gillespie, Daniel T. (1977). Exact Stochastic


simulation of coupled chemical reactions. Journal of
Physical Chemistry, 81 (25), 23402361.

[6]

Gillespie, Daniel T. (1992). A rigorous derivation of


the chemical master equation. Physica A, 188, 404425.

ISBN: 978-1-941968-35-2 2016 SDIWC

91

Potrebbero piacerti anche