NetDS: An OpenCL-based Cytoscape plugin for fast parallel analysis of robustness dynamics and feed-forward/feedback loop structures in large-scale networks

ruralrompΛογισμικό & κατασκευή λογ/κού

2 Δεκ 2013 (πριν από 4 χρόνια και 29 μέρες)

60 εμφανίσεις


-

1

-

Systems biology

NetDS
par
: An OpenCL
-
based Cytoscape plugin for fast parallel analysis of
robustness
dynamics
and feed
-
forward/feedback loop structures

in large
-
scale
networks

Hung
-
Cuong Trinh
1
, Duc
-
Hau Le
2

and Yung
-
Keun Kwon
1,
*

1
School of
Electrical Engine
ering
, University of Ulsan,
93, Daehak
-
ro,
Nam
-
gu, Ulsan 680
-
749.

2
School of Computer Science and Engineering, Water Resources University, 175 Tay

Son, Dong Da, Hanoi,
Vietnam.



User Manual

I.

How
to setup

................................
................................
................................
................................

2

II.

How to use

................................
................................
................................
................................
...

4

1.

Run NetDSpar plugin

................................
................................
................................
................

5

a.

Load Net
DSpar control panels

................................
................................
..............................

5

b.

Setup OpenCL options

................................
................................
................................
..........

6

2.

Network Creation

................................
................................
................................
......................

7

a.

Random network generation and simulation

................................
................................
.........

7

b.

Import network

................................
................................
................................
......................

8

3.

Set and perturb update
-
rules and states of nodes

................................
................................
....

10

4.

Network dynamics analysis

................................
................................
................................
....

11

a.

State Transition
................................
................................
................................
....................

12

b.

Robust
ness Examination

................................
................................
................................
.....

12

5.

Network structure analysis

................................
................................
................................
......

14

a.

Analysis of
Feedback loops

................................
................................
................................
.

14

b.

Analysis of f
eed
-
forward loops

................................
................................
...........................

15

6.

Relationships between structural and dynamic properties

................................
......................

17

a.

Case

study 1:

................................
................................
................................
.......................

17

b.

Case study 2:

................................
................................
................................
.......................

17

c.

Case study 3:

................................
................................
................................
.......................

17

d.

Case stu
dy 4:

................................
................................
................................
.......................

18

e.

Case study 5:

................................
................................
................................
.......................

19



-

2

-


I.

How to setup

-

Copy jar file (NetDSpar.jar) to
plugins

folder in the installed folder of Cytoscape.

-

OpenCL setup:

o

Download "
Accel
erated Parallel Processing (APP) SDK
" from AMD website
http://developer.amd.com/

(recommended version is v2.8), and insta
ll it.

o

Install appropriate drivers for all Graphics cards of the computer.


-

Download example network files from
http://netdspar.sourceforge.net/

to see file format and test.

-

Set the maximum heap size by open
ing the file
Cytoscape.vmoptions

(which is located in the
installed folder of Cytoscape) and modify the parameter "
-
Xmx
" , for example:

-
Xmx1548m

o

If the computer has a large memory size, we'd better increase the heap size, such as in a
computer with 8G mem
ory, we can set:



-
Xmx6548m

o

Note that in Windows 7, we must run the editor software (for ex. Notepad) with
Administrator permission to modify this file.


-

3

-


-

Note:

Run the Cytoscape software with Administrator permission (for ex., in Windows 7 we can
right
-
clic
k on the icon of Cytoscape and choose "
Run as administrator
"), because NetDSpar needs
permission to write result files in some functions.


-

4

-


II.

How to use

We demonstrate with some case studies based on following networks:

-

CDRN

(Use for section 2, 3, 4, 5
, 6
)

o

T
he cell differentiation regulatory

network with 9 nodes and 15 links

(
Huang

et al.
,
2000
).

o

Use the file “
CDRN
-

Cell differentiation regulatory network.txt


-

AMRN

(Use for section
6
)

o

The
Arabidopsis

morphogenesis regulatory network

with 10 nodes and 22 links
(
Mendoza

et al.
, 1999
).

o

Use the
file

AMRN
-

Flower morphogenesis in Arabidopsis.txt


-

YCCN

(Use for section
6
)

o

The yeast cell cycle network with 11 nodes and 34 links (
Li

et al.
, 2004
);

o

Use the file “
YCCN
-

Yeast Cell
-
cycle network.txt
”.

-

HSN

(Use for section
6
)

o

The

large
-
scale human signaling network
(HSN) with 1609 nodes and 5063

links (
Cui

et
al.
, 2009
)

o

Use the
file


HSN
-

Human signaling network.txt



Usage of
NetDSpar

will be demonstrated by following s
ections:

1.

Run
NetDSpar

plugin

2.

Network Creation

3.

Set and perturb
update
-
rule
s and states of nodes

4.

Network dynamics analysis

5.

Network structure analysis

6.

Relationships between structural and dynamic properties


-

5

-


1.

Run
NetDSpar

plugin

a.

Load
NetDSpar

control panels

-

Run Cytoscape,
NetDSpar

plugin will be automatically loaded in Plugins menu of Cytoscape as
following:




-

Choose
Network Dynamics & Structure

-

NetDSpar

has two main control panels:

o

One for
Network dynamics analysis


o

One for
Network structure analysis



-

6

-

b.

S
etup OpenCL options

-

Choose
Setting...

menu

-

OpenCL Setting menu has three main options:

o

CPU with one core
: only use one core of the CPU for tasks



With this option, NetDSpar is equivalent to original NetDS.

o

OpenCL on CPU multi
-
core
: use all cores of the CP
U for parallelization
of tasks

o

OpenCL on GPU device
: use GPU device for parallelization of tasks



-

7

-


2.

N
etwork

Creation

a.

Random network

generation
and

simulation




-

Choose menu
Plugin/
NetDSpar
/Random Network
Generation
&
Simulation
...

-

Have
three

main panels:

o

Step 1:
Random Boolean
Network generation:



Choose one of random Boolean network model
s




Barabási
-
Albert model (
Barabasi

et al.
, 1999
),



E
r
dős
-
Rényi model (
Erdös

et al.
, 1959
),



Erdős
-
Rényi variant model



Shuffling model



Choose parameters to generate corresponding random Boolean networks



For
the first
three

models,
users can set
the probability of negative link's
assignment (
this probability would affect
the ratio of negative links

to all links)
.



For Shuffling model:

create random networks
from a given network.



Choose "
Shuffle direction and sign of all interactions
" if users want
to
create random networks by shuffling the direction and the sign of
every interaction from the

given ne
twork
: Shuffle I
.



Choose "
Preserve in
-
degree and out
-
degree of all nodes
" if users
want to create random networks by

rewiring the
edges

of the given
network such that
the in
-
degree and
the
out
-
de
gree of all nodes are
conserved
: Shuffle II
.

-

Users can set th
e value of "
Shuffling
intensity
"
parameter.
The
number of rewiring steps
=

Shuffling intensity

×

(number of
edges)
.




Choose
"
Don't create view for random networks"

if users don't want to
create view for random networks when the simulation is running.

If th
is option
is selected, NetDSpar can save
a significant amount of

memory.


-

8

-

o

Network dynamics
:

c
alculate robustness against initial
-
state perturbation and
update
-
rule perturbation over a chosen set of initial
-
states



Choose a update
-
rule scheme for each node of

all r
andom networks:



CONJ and DISJ: denote that each node of a
random network

would be assigned a conjunction and disjunction function,
respectively.



CONJ
-
DISJ
:

denotes that each node of a
random network

would
be assigned a conjunction or disjunction fun
ction randomly
.



Choose “
Robustness against Initial
-
state mutation
” if users want
to calculate
robustness against initial
-
state perturbation
value
of each node of all
random networks
.



Choose “
Robustness against
Update
-
rule

mutation
” if users want
to calcula
te
robustness against
update
-
rule

perturbation value of each node of all
random networks
.

o

Network's topological properties:

there are two
subsections related to Feedback/Feed
-
forward loop analysis



Feedback & Coupled
Feedback L
oop A
nalysis:



Choose the leng
th of feedback loops to be returned
.



Choose “
Less than or Equal to
” if users want all FBLs with the length
from 2 to specified length are retrieved.



Choose “
Find Coupled Feedback L
oops
” if users want to check
whether all retrieved feedback loops are couple
d.



Feed
-
forward

L
oop A
nalysis:



Choose the length of paths to be returned
.



Choose “
Less than or Equal to
” if users want all paths with the length
from 1 to specified length are retrieved.

-

Have an option for saving results:

o

Choose “
Save detail
ed

results for
each node of all networks
” if users want
to save
two types of robustness and FBLs/FFLs results of each node of all random
networks
.

Otherwise,
only
the number of positive/negative FBLs,
the
number of
coherent/incoherent FBLs/FFLs for each random network

we
re saved.



b.

Import network

-

Networks which are imported can be pre
-
generated RBN, signaling or regulatory network.

o

Note that
:




T
o use imported networks in this plugin,

the networks file must have at least 3
columns, source node, interaction type and target
node

where the

interaction
type only can be one of
-
1, 1 and 0 that are corresponding to inhibition,
activation and neutral, respectively. For example
,
CDRN

network:


-

By using

the
b
uilt
-
in import function of Cytoscape
,

users can select file in some format
s and
users have to specify which column is source, target and interaction type
.


-

9

-




For example:
CDRN network

(
Huang

et al.
, 2000
).
Choose “
Organic
” from
Layout/
yFiles

menu





-

10

-


3.

Set and perturb
update
-
rule
s and states of nodes

Each node in the network has its own state and
update
-
rule
.



Right after network is either generated or imported,
update
-
rule
s and states of all nodes are
randomly assigned, but then users
can specify those manually.



With
update
-
rule
s of nodes, we provided three updating
-
rule schemes; CONJ
-
DISJ, CONJ, and
DISJ.

o

If CONJ
-
DISJ is chosen, then users can either specify or randomize
update
-
rule

for each
node.

o

Note that, 1 and 0 are denoted disj
unction and conjunction, respectively.



U
sers also can either specify or randomize state for each node.

o

Note that
, 1 and 0 are denoted “on” or “off” state of each node, respectively.


In visualization



N
odes having state of 1 and 0 are colored in gray and

light
-
gray, respectively.



N
odes having
update
-
rule

of 1 and 0 are in rectangle and circle shape, respectively.

The state and
update
-
rule

of each node is also stored as attributes of each node, and they are updated
whenever they are changed either by set
ting
values in
Node Attribute Browser

Tab manually

or
simulation of network dynamics. F
ollowing f
igure showed simulations of these issues for the
CDRN
.


Node Attribute Browser




-

11

-


4.

Network dynamics analysis

Control panel



Network dynamic can be examine
d based on a set of initial state. Here, we provided 3 options:


Note that:


-

W
hen the network size is large (large
-
scale network) (i.e., number of nodes >20) then
“Over # of random States” should be chosen. The outcome of analyses is similar to that
of th
e option “Over All Possible States”

(See following figures. Number of initial states
is only 200 (all possible state is 512), but number of attractors return
ed

is same (6
attractors)
)
.


-

12

-


a.

State Transition

Control panel



-

Select “
Attractors Only
” if user wa
nt only to see attractors. For example. Network transition
over 3 above options:









-

User can also check which state belongs to attractor by selecting state transition network and see
“Attractor” attribute as following (1 means a state belongs

to an attractor):


b.

Robustness Examination

Control panel


-

Choose type of perturbation (mutation) against which the network/node robustness is examined.

-

Choose nodes in the network

-

Choose
set

of initial states


Note that:


-

13

-

-

If “Over the Specified State” op
tion is selected. Then a
selected
node will be decided as
“Robust” (highlighted in
green

after examining) or “Not Robust”

(highlighted in
red

after examining)
.

-

If “Over # of random States” or “Over All Possible States” option is selected. Then a

selected

n
ode is

given a robustness value which is calculated over chosen set of

initial
states.

o

T
he colo
u
r of
a node depends on
its robustness value

(ranged between 0 and
1)

as following gradient colo
u
r

bar

-

the node with robustness value 0 has
Red colour and the
node with robustness value 1 has Green colour:


-

If “Over All Possible States” option is selected and user wants to examine Robustness,
then user is asked to choose whether all possible update
-
rule combinations are examined
or not. If “Yes” then the robust
ness of nodes and network is average over all update
-
rule
combinations.
Details of robustness of each node for each update
-
rule combination are

store
d

in file “Summary_RobustnessOfNodes.txt” in Cytoscape installed folder.


-

If all nodes in the network are
selected, then the robustness of network will be
calculated.

-

Robustness of network is also stored as network attribute

in
Network Attribute
Browser
.

-

Robustness of nodes is also stored as node attribute. It is important for further process
es

such as statist
ics and plot charts.



-

“Over the Specified State”
潰oi潮⁩猠sele捴ed

o潢畳on敳猠潦⁥慣栠h潤o



o潢畳on敳s 潦 湥tw潲k ag慩湳n two
ty灥p ⁰敲t畲扡bi潮



-

All潤 猠sr攠e敬e捴ed

-

“Over All Possible States” option is
獥le捴敤e

-

A 獰s捩fi敤eu灤pte
-
r畬
攠c潭扩湡nion
i猠sx慭i湥搮

o潢畳on敳猠潦⁥慣栠h潤o



-

All潤 猠sr攠e敬e捴ed

-

“Over All Possible States” option
i猠s敬e捴敤e

-

All⁵灤 te
-
r畬攠e潭扩湡ni潮猠or攠
數emi湥搮







-

14

-


5.

Network structure analysis

NetDSpar

can find Feedback, Feed
-
forward loops,
paths, check whether loops are coupled and decide they
are coherent or incoherent.



a.

Analysis of
Feedback loops

Control Panel


-

Select a set of nodes in the network

-

Choose the length of feedback loops to be returned

-

Choose “
Less than or Equal to
” if users

want all FBLs with the length from 2 to specified
length are retrieved.

-

Choose “
Show distinct Feedback loops
” if users want to see only all distinct feedback loops
involved in selected nodes.

-

Choose “
Find Coupled Feedback loops
” if users want to check whe
ther all retrieved feedback
loops are coupled. A list of coupled feedback loops are shown in “
Coupled Feedback Loop
List
” Tab.


Note that:

-

Number of FBLs (Number of positive FBLs, negative FBLs and all) are also calculated
and stored as node attribute. It
is important for further process
es

such as statistics and
plot charts.

-

If users choose all nodes in the network

and “
Less than or Equal to

, then number of
FBLs in the network with
the length from 2 to specified length are retrieved and stored
as network a
ttribute.

Also,
numbers of coherent and incoherent coupled FBLs of
network are

stored as network attribute.







-

15

-

-

FBLs with detail information as Type, nodes in it are shown in each row.

-

Select one or more to highlight them in the network or press

“Show” button to display in separate window.




-

Number of FBLs is stored as node attribute.




-

All coupled FBLs are checked. The first column denotes the type of coupled FBL, the second column
denotes the intersection length and nodes in the int
ersection between two FBLs, the third column reveals the
type of FBL involved in a selected coupled FBL, and the last columns denote the nodes in each FBL.

-

Select 2 consecutive odd and even rows to view a coupled FBL. A selected coherent coupled FBL is
shown
in a separate window.


b.

Analysis of f
eed
-
forward loops

Control Panel


-

Select pairs of nodes (“
From
” and “
To
”)

-

Choose the length of paths to be returned

-

Choose “
Less than or Equal to
” if users want all paths with the length from 1 to specified
length

are retrieved.

-

Choose “
Find Feed
-
forward Loops
” if users want to find FFL. A list of FFLs are shown in

Feed
-
forward Loop List
” Tab.


Note that:

-

If user choose all nodes in “
From
” and all nodes in “
To
”, then all paths in the network are
retrieved. And if


Find Feed
-
forward Loops
” is chosen, then all FFLs of the network will be
found and stored as network attribute.




-

16

-



-

A total of 178 simple paths with length


8 were found in the CDRN when all possible pairs of nodes were
selected (left panel). A s
elected path is shown in a separate window (right panel)





-

A total of
64

FFLs were found including
49

coherent and 15

incoherent FFLs (left panel). These figures were stored
as network attributes.

-

The first column denotes the type of FFL, the se
cond column denotes the input and output nodes of the FFL, the
third column shows the type of each path, and the last columns denote the nodes belonging to each path.

-

A coherent FFL is shown in a separate window (right panel).



-

17

-


6.

Relationships between
structural and dynamic properties

a.

Case study 1:

R
obust networks tend to have a larger number of positive FBLs and a smaller number of negative
FBLs (
Kwon

et al.
, 2008b
). This is partially supported by previous research. For example:

-

AMRN
has four positive and two negative

FBLs. This regulatory network is known to robustly
control the process of flower development (
Mendoza

et al.
, 1999
).

-

CDRN has

seven positive and two negative FBLs is found to robustly induce quiescence,
terminal differentiation, and apoptosis (
Huang

et al.
, 2000
).

-

HSN
-

human signaling network
(HSN) (
Cu
i

et al.
, 2009
)

Following table shows the number of positive and negative FBLs of those networks:



Number of
positive FBLs

Number of
negative FBLs

AMRN

(Maximal Length =10)

4

2

CDRN


(Maximal Length =9)

7

2

HSN


(Maximal Length =
8
)

445434

273600


b.

Cas
e study 2:

A previous study showing that the coherent coupling of FBLs is a design principle of a cell signaling
network (
Kwon

et al.
, 2008a
). More specifically, a larger number of coherent coupled FBLs than
incoherent coupled FBLs were found in the cell signaling network and this

strengthened the robustness
of the network.

For example:
With

CDRN, HSN,
and
YCCN
.

We found that all of the
them

have a larger number of
coherent FBLs than that of incoherent FBLs

as
following table:




Number of
Coherent FBLs

Number of
Incoherent FBLs

YCCN


(Maximal Length =11)

473

337

CDRN


(Maximal Length =9)

14

6

HSN


(Maximal Length =5
)

308816

160992


c.

Case study 3:

A previous study showing that the number of FBLs is negatively correlated with the robustness of
network nodes was examined (
Kwon

et al.
, 2007b
). Here, the relationship between the number of FBLs
and
γ
s
(
v
) over all possible initial states in the CDRN was investigated.

-

To this end, the average
γ
s
(
v
) over all possible update
-
rules (512 combinations in total) was
calculated.

o

After finding robustness of all nodes in the network over all possible states an
d all
update
-
rule combinations, use robustness
summary
of nodes over each update
-
rule
combination in file “
Summary_RobustnessOfNodes.txt
” in Cytoscape installed folder.

o

Find all FBLs of all nodes in the networks. Collect number of FBLs for each node in
N
o
de
A
ttribute
Browser
.

o

Draw chart based on robustness and number of FBLs of nodes

-

We detected a negative relationship between the number of FBLs and
γ
s
(
v
) (the slope of the
regression line is
-
0.02288 and the P
-
value


0.00405).


-

18

-


d.

Case study 4:

A

study showing that networks tend to have a larger proportion of basins belonging to fixed
-
point
attractors as they have more positive FBLs (
Gouze, 1998; Kwon

et al.
, 2007a
) was validated.

-

Examine the basins of fixed
-
point and limit
-
cycle attractors in the
CDRN

and
A
MRN
.

o

Using the CONJ scheme,

o

Load networks to Cytoscape workspace.

o

Examine state transition over all possible states for each network.

o

Count number of states which belongs to attractor by choosing state transition network
and looking at the node attribute

“Attractor” (1 means this state belongs to an attractor).

-

Find all FBLs of networks. Go to
N
etwork
Attribute Browser
to see number of positive and
negative FBLs of the network.

As a result:

-

In the
CDRN
,
5

fixed
-
point attractors and
1

limit
-
cycle attracto
r were found; the basin sizes of
the fixed
-
point and limit
-
cycle attractors were 488 and 24, respectively.

-

In the
AMRN
, 26 fixed
-
point attractors and 8 limit
-
cycle attractors were found; the basin sizes of
the fixed
-
point and limit
-
cycle attractors were 8
32 and 192, respectively.

-

Both networks have a larger number of positive FBLs than negative FBLs, as shown in
following f
igure. Thus, the results support the hypothesis regarding the positive relationship
between the basin of fixed
-
point attractors and th
e number of positive FBLs.

CDRN



Attractor

type

Number of
attractors

Basin

size

Fixed
-
point

5

488/2
9

Limit
-
cycle

1

24/2
9


AMRN




Attractor

type

Number of
attractors

Basin

size

Fixed
-
point

26

832/2
10

Limit
-
cycle

8

192/2
10





-

19

-

e.

Case study 5:

In this

section
, we present a case study for the new function
of
NetDS
par
: batch
-
mode simulation on
RBNs
.
Two
previous stud
ies

showing that

coherent FBLs/
FFLs
are ubiquitously found

in
cell signaling
networks

and
explained that abundant coherent
FBL
s

and coherent

FFLs are needed to
strengthen

the
robustness against
initial
-
state perturbations and
update
-
rule perturbations
, respectively
.

Here
, we use
the
batch
-
mode function
to
validate the abundance of coherent FBLs/FFLs in t
he HSN.

First
,

the network file of HSN
i
s needed to

be
imported

into Cytoscape software.
Next,

we execute
two

RBNs simulations

by using the batch
-
mode function.
To this end, we
choose

shuffling technique
s

for

the

simulation
s
: Shuffle I
for the first simulation
and Shuffle II

for the second simul
ation

("
Shuffling
intensity
" parameter is
set

to 4)
. For each
simulation
, we set
the number of

random networks
t
o 1
000
,
check all options of FBLs/FFLs section
s
,

and
set

the length of FB
Ls and paths to
5 and 2 respectively.
We recommend users to
setup OpenC
L options for
faster simulations

by the parallelization of searching
FBLs/FFLs, and also check the option
"Don't create view for random networks"

to save a significant
amount of memory
.

Finally, w
e compare the ratio
s

NuCoFBL/
NuCoupledFBL
and
NuCoFFL/NuFFL
of the real
biological network HSN with
those

of random networks generated by two

the

shuffling techniques. As in
following t
able
s
, the ratio
s

NuCoFBL/NuCoupledFBL
and
NuCoFFL/NuFFL
in the human signaling
network
are

significantly greater than
those

of the

shuffled random networks for both two shuffling
algorithms (using one
-
sample t
-
test, P
-
value<0.0001). This indicates that coherent FBLs
/FFLs

are
ubiquitous in the large
-
scale human signaling network.




Number of
Coherent FBLs

(NuCoFBL)

Number of
Incohere
nt FBLs

(NuInCoFBL)

NuCoFBL/

NuCoupledFBL

HSN

Maximal
Length =3

899

493

0.64583

Maximal
Length =4

16600

8595

0.65886

Maximal
Length =5

308816

160992

0.65732

RBNs
(Shuffle I)

Maximal
Length =5



µ


〮0〰㘴

E
σ ≈

〮〰〷0
F

o_k猠
E卨畦fl攠efF

䵡硩m慬
䱥i
gt栠㴵






〮0㜵㤰

E
σ ≈

〮〱㜲T
F

Note
:

NuCoupledFBL

= NuCoFBL + NuInCoFBL
;
µ and
σ

ar攠e敡渠慮n⁳ 慮摡rd
摥di慴i潮o敲‱ 〰⁒_k猠






k畭扥b
C潨敲敮t⁆ 䱳

Ek畃潆o䰩

k畭扥b
f湣潨nr敮e⁆ 䱳

Ek畉湃潆o䰩

k畃潆o䰯
k畆ui

e华

䵡硩m慬
䱥湧t栠㴲

4417

1
455

0.75221

RBNs
(Shuffle I)

Maximal
Length =2



µ


〮0㔳㘶

E
σ ≈

〮〱〲0
F

o_k猠
E卨畦fl攠efF

䵡硩m慬
䱥湧t栠㴲






〮0㌱㠹

E
σ ≈

〮〱㌲P
F

Note
:

NuFFL

=
NuCoFFL

+
NuInCoFFL
;
µ and
σ

慲攠m敡渠慮搠nta湤nr搠摥diatio渠潶敲
㄰〰⁒_ks




-

20

-

References

Barabasi, A.L. and Albert, R. (1999) Emergence of Scaling in Random Networks,
Science
,
286
, 509
-
512.

Cui, Q.

et al.

(2009) Protein evolution on a human signaling network,
BMC Systems Biology
,
3
, 21.

Erdös, P. and Rényi, A. (1959) On random graph
s, I,
Publicationes Mathematicae (Debrecen)
,
6
, 290
-
297.

Gouze, J.
-
L. (1998) Positive and negative circuits in dynamical systems,
Journal of Biological Systems
,
6
,
11

15.

Huang, S. and Ingber, D.E. (2000) Shape
-
Dependent Control of Cell Growth, Differentia
tion, and Apoptosis:
Switching between Attractors in Cell Regulatory Networks,
Experimental Cell Research
,
261
, 91
-
103.

Kwon, Y.
-
K. and Cho, K.
-
H. (2007a) Boolean Dynamics of Biological Networks with Multiple Coupled
Feedback Loops,
Biophysical Journal
,
92
, 2975
-
2981.

Kwon, Y.
-
K. and Cho, K.
-
H. (2008a) Coherent coupling of feedback loops: a design principle of cell
signaling networks,
Bioinformatics
,
24
, 1926
-
1932.

Kwon, Y.
-
K. and Cho, K.
-
H. (2008b) Quantitative analysis of robustness and fragility in biolo
gical networks
based on feedback dynamics,
Bioinformatics
,
24
, 987
-
994.

Kwon, Y.
-
K.

et al.

(2007b) Investigations into the relationship between feedback loops and functional
importance of a signal transduction network based on Boolean network modeling,
BMC

Bioinformatics
,
8
, 384.

Li, F.

et al.

(2004) The yeast cell
-
cycle network is robustly designed,
Proceedings of the National Academy
of Sciences of the United States of America
,
101
, 4781
-
4786.

Mendoza, L.

et al.

(1999) Genetic control of flower morphogenesis in Arabidopsis thaliana: a logical
analysis,
Bioinformatics
,
15
, 593
-
606.