4. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Partially NCFs
De
5. nitions
Depth and stability
Derrida plots
Conclusions
Elena Dimitrova Canalyzation in mathematical modeling
6. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Networks
Elena Dimitrova Canalyzation in mathematical modeling
7. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
What is a Boolean network?
Elena Dimitrova Canalyzation in mathematical modeling
8. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
What is a Boolean function?
I Boolean function (or switching function) is a function of the
form f : Bn ! B, where B = f0; 1g, n is nonnegative integer.
I Boolean functions are often written using the logical operators
AND (^), OR (_), and NOT (:).
Elena Dimitrova Canalyzation in mathematical modeling
9. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Boolean function: Example
f (x; y) = x ^ :y = x(y + 1)
x y f
0 0 0
0 1 0
1 0 1
1 1 0
Facts
I Every Boolean function can be represented as a polynomial
over F2.
I Every function f : Fn2
! F2 can be represented as a Boolean
function.
Elena Dimitrova Canalyzation in mathematical modeling
10. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
What is a Boolean network?
I F = (f1; : : : ; fn) : Bn ! Bn is a Boolean network (BN) on n
variables if fi is a Boolean function in n variables for every
i = 1; : : : ; n.
I BNs are used for modeling networks in which the node
activity/state can be described as a binary value: on-o,
active-non active, 1-0, etc.
Elena Dimitrova Canalyzation in mathematical modeling
12. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Applicatioins of Boolean networks
Boolean network models have been used to examine the
connections among diverse physical and engineered networks
I Signal transduction networks [Kauman, Shmulevich et al.,
Helikar et al., Kochi et al.]
I Biological networks [Klemm et al., Raeymaekers]
I Social networks [Flache et al., Green et al., Moreira et al.]
I Economic/prediction market networks [Jumadinova et al.]
I Neural networks [Huepe et al.]
I Complex networks in general [Wolfram]
Elena Dimitrova Canalyzation in mathematical modeling
13. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Boolean networks: Biology
Segment polarity network in D. melanogaster
R. Albert H. Othmer: J. Theor. Bio. 223 (2003), 1{18.
Elena Dimitrova Canalyzation in mathematical modeling
14. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Boolean model of segment polarity network
R. Albert H. Othmer: J. Theor. Bio. 223 (2003), 1{18.
Elena Dimitrova Canalyzation in mathematical modeling
15. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Boolean networks: Electric circuits
Elena Dimitrova Canalyzation in mathematical modeling
16. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Boolean networks: Sociology
I Databases like Citebase harvest pre- and post-prints from
various archives for dierent
17. elds. Nodes stand for published
articles and a directed edge represents a reference to a
previously published article.
I The International Movie Database is analyzed in many
studies. An actor collaboration network can be reconstructed,
in which actors casted jointly are assumed to know each other.
I The phenomenon of rhythms of social interactions was studied
using Boolean networks. A clear weekend pattern was
discovered while studying messaging within a massive online
friendship site providing insight to the social life of the users.
(www.hpl.hp.com=research=idlpapers=facebook)
Elena Dimitrova Canalyzation in mathematical modeling
18. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Variations on Boolean networks
I (Random/Probabilistic) Boolean networks (S. Kauman)
I Polynomial dynamical systems (T. Thomas, D. Thiery, R.
Laubenbacher, B. Stigler, B. Pareigis, A. Jarrah)
I (Nested) canalyzing functions (S. Kauman, I. Shmulevich)
I Cellular automata (von Neumann, J. Conway, S. Wolfram)
I Graphical models (Barret, Mortveit, Reidys)
I Logical models (R. Thomas)
I Petri nets (Google `Petri Nets World')
I Biologically meaningful functions (L. Raeymaekers)
I ...
Elena Dimitrova Canalyzation in mathematical modeling
19. Boolean networks
Canalyzing functions
Nested canalyzing functions
Partially NCFs
Question
Are all Boolean functions good for modeling??
Elena Dimitrova Canalyzation in mathematical modeling
21. nitions
Examples and applications
Canalyzing functions
Canalyzing function
A Boolean function f (x1; : : : ; xn) is canalyzing if there exists an
index i and a; b 2 f0; 1g such that
f (x1; : : : ; xi1; a; xi+1; : : : ; xn) = b:
That is, when xi is given the input value a, f evaluates to b
regardless of the input values of the other variables. Here a is
called the canalyzing value and b is called the canalyzed value.
Elena Dimitrova Canalyzation in mathematical modeling
23. nitions
Examples and applications
Examples: Canalyzing function
I The AND function x ^ y is canalyzing in x. Since 0 ^ y = 0
for any input value of y, 0 is a canalyzing value for x with
canalyzed output value 0.
I XOR(x; y) := (x _ y) ^ :(x ^ y) is not canalyzing in either
variable.
Elena Dimitrova Canalyzation in mathematical modeling
25. nitions
Examples and applications
Examples: Canalyzing function
I The AND function x ^ y is canalyzing in x. Since 0 ^ y = 0
for any input value of y, 0 is a canalyzing value for x with
canalyzed output value 0.
I XOR(x; y) := (x _ y) ^ :(x ^ y) is not canalyzing in either
variable.
Elena Dimitrova Canalyzation in mathematical modeling
27. nitions
Examples and applications
Canalyzing functions: role and applications
I Introduced by Kauman [Kau93] as appropriate rules in
Boolean network models of gene regulatory networks. The
de
28. nition is reminiscent of the concept of canalisation
introduced by the geneticist C.H. Waddington [Wad42] to
represent the ability of a genotype to produce the same
phenotype regardless of environmental variability.
I Play an important role in the study of random Boolean
networks [Kau93, Lyn95, Sta87, Ste99].
I Have been used extensively as models for dynamical systems,
such as gene regulatory networks [Kau93], evolution [Ste99],
and chaos [Lyn95].
Elena Dimitrova Canalyzation in mathematical modeling
30. nitions
Examples and applications
Canalyzing functions: role and applications
I Introduced by Kauman [Kau93] as appropriate rules in
Boolean network models of gene regulatory networks. The
de
31. nition is reminiscent of the concept of canalisation
introduced by the geneticist C.H. Waddington [Wad42] to
represent the ability of a genotype to produce the same
phenotype regardless of environmental variability.
I Play an important role in the study of random Boolean
networks [Kau93, Lyn95, Sta87, Ste99].
I Have been used extensively as models for dynamical systems,
such as gene regulatory networks [Kau93], evolution [Ste99],
and chaos [Lyn95].
Elena Dimitrova Canalyzation in mathematical modeling
33. nitions
Examples and applications
Canalyzing functions: role and applications
I Introduced by Kauman [Kau93] as appropriate rules in
Boolean network models of gene regulatory networks. The
de
34. nition is reminiscent of the concept of canalisation
introduced by the geneticist C.H. Waddington [Wad42] to
represent the ability of a genotype to produce the same
phenotype regardless of environmental variability.
I Play an important role in the study of random Boolean
networks [Kau93, Lyn95, Sta87, Ste99].
I Have been used extensively as models for dynamical systems,
such as gene regulatory networks [Kau93], evolution [Ste99],
and chaos [Lyn95].
Elena Dimitrova Canalyzation in mathematical modeling
38. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Nested Canalyzing Functions
Let f (x1; : : : ; xn) be a Boolean function. For 2 Sn, f is a nested
canalyzing function (NCF) in the variable order x(1); : : : ; x(n)
with canalyzing values a1; : : : ; an and canalyzed values b1; : : : ; bn if
it can be expressed in the form
f =
8
:
b1 if x(1) = a1
b2 if x(1)6= a1 and x(2) = a2
b3 if x(1)6= a1 and x(2)6= a2 and x(3) = a3
...
...
bn if x(1)6= a1 and and x(n1)6= an1 and x(n) = an
:bn if x(1)6= a1 and and x(n)6= an
Elena Dimitrova Canalyzation in mathematical modeling
40. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Examples: NCFs
I f (x1; x2; x3) = x2 _ (:x1 ^ x3) is a NCF with b = [1; 0; 0],
a = [1; 1; 0], and = [2; 1; 3]:
f (x1; x2; x3) =
8
:
1 if x2 = 1
0 if x2 = 0 and x1 = 1
0 if x2 = 0, x1 = 0, and x3 = 0
1 if x2 = 0, x1 = 0, and x3 = 1:
I f (x; y; z) = x(y 1)z is a NCF.
I g(x; y; z;w) = x(y + z) is not a NCF.
Elena Dimitrova Canalyzation in mathematical modeling
42. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Examples: NCFs
I f (x1; x2; x3) = x2 _ (:x1 ^ x3) is a NCF with b = [1; 0; 0],
a = [1; 1; 0], and = [2; 1; 3]:
f (x1; x2; x3) =
8
:
1 if x2 = 1
0 if x2 = 0 and x1 = 1
0 if x2 = 0, x1 = 0, and x3 = 0
1 if x2 = 0, x1 = 0, and x3 = 1:
I f (x; y; z) = x(y 1)z is a NCF.
I g(x; y; z;w) = x(y + z) is not a NCF.
Elena Dimitrova Canalyzation in mathematical modeling
44. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Examples: NCFs
I f (x1; x2; x3) = x2 _ (:x1 ^ x3) is a NCF with b = [1; 0; 0],
a = [1; 1; 0], and = [2; 1; 3]:
f (x1; x2; x3) =
8
:
1 if x2 = 1
0 if x2 = 0 and x1 = 1
0 if x2 = 0, x1 = 0, and x3 = 0
1 if x2 = 0, x1 = 0, and x3 = 1:
I f (x; y; z) = x(y 1)z is a NCF.
I g(x; y; z;w) = x(y + z) is not a NCF.
Elena Dimitrova Canalyzation in mathematical modeling
46. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Properties of NCFs
I Introduced by Kauman et al. in [KPST03].
I [KPST04] showed that networks made from NCFs have stable
dynamic behavior and might be a good class of functions to
express regulatory relationships in biochemical networks.
I Nested (hierarchically) canalyzing functions show ordered
behavior [NFW07].
I Systems of NCFs have a smaller average cycle length and
average height (number of time steps it takes to converge to
an attractor) of the state space graph compare to general
Boolean networks.
Elena Dimitrova Canalyzation in mathematical modeling
48. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Properties of NCFs
I Introduced by Kauman et al. in [KPST03].
I [KPST04] showed that networks made from NCFs have stable
dynamic behavior and might be a good class of functions to
express regulatory relationships in biochemical networks.
I Nested (hierarchically) canalyzing functions show ordered
behavior [NFW07].
I Systems of NCFs have a smaller average cycle length and
average height (number of time steps it takes to converge to
an attractor) of the state space graph compare to general
Boolean networks.
Elena Dimitrova Canalyzation in mathematical modeling
50. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Properties of NCFs
I Introduced by Kauman et al. in [KPST03].
I [KPST04] showed that networks made from NCFs have stable
dynamic behavior and might be a good class of functions to
express regulatory relationships in biochemical networks.
I Nested (hierarchically) canalyzing functions show ordered
behavior [NFW07].
I Systems of NCFs have a smaller average cycle length and
average height (number of time steps it takes to converge to
an attractor) of the state space graph compare to general
Boolean networks.
Elena Dimitrova Canalyzation in mathematical modeling
52. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Properties of NCFs
I Introduced by Kauman et al. in [KPST03].
I [KPST04] showed that networks made from NCFs have stable
dynamic behavior and might be a good class of functions to
express regulatory relationships in biochemical networks.
I Nested (hierarchically) canalyzing functions show ordered
behavior [NFW07].
I Systems of NCFs have a smaller average cycle length and
average height (number of time steps it takes to converge to
an attractor) of the state space graph compare to general
Boolean networks.
Elena Dimitrova Canalyzation in mathematical modeling
54. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Representation of an NCF [JRL07]
Let yi = x(i) + ai + bi , 1 i n and let
g(x1; : : : ; xn) = y11(y22(: : : (yn1n1yn) : : :);
where
i =
_ if bi = 1
^ if bi = 0
and ai ; bi 2 f0; 1g for 1 i n. Then g is an NCF in the variable
order x(1); : : : ; x(n) with canalyzing values a1; : : : ; an and
canalyzed values b1; : : : ; bn. Further, any NCF can be represented
in this form.
Elena Dimitrova Canalyzation in mathematical modeling
56. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
NCFs as a toric variety
I The ring of Boolean functions is isomorphic to the ring
R = F2[x1; : : : ; xn]=hx2
i xi : 1 i ni.
I Any polynomial in R can be represented as an point in F2n
2
corresponding to its coecients.
I For 2 Sn, the set of points corresponding to NCFs in the
variable order x(1); : : : x(n) form an algebraic variety VNCF
.
I VNCF = [2SnVNCF
is an algebraic variety as well, the
variety of nested canalyzing functions.
I VNCF is a toric variety [JL07].
Elena Dimitrova Canalyzation in mathematical modeling
58. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
NCFs as a toric variety
I The ring of Boolean functions is isomorphic to the ring
R = F2[x1; : : : ; xn]=hx2
i xi : 1 i ni.
I Any polynomial in R can be represented as an point in F2n
2
corresponding to its coecients.
I For 2 Sn, the set of points corresponding to NCFs in the
variable order x(1); : : : x(n) form an algebraic variety VNCF
.
I VNCF = [2SnVNCF
is an algebraic variety as well, the
variety of nested canalyzing functions.
I VNCF is a toric variety [JL07].
Elena Dimitrova Canalyzation in mathematical modeling
60. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Binary decision diagrams
I A binary decision diagram (BDD) is a directed, acyclic graph
that corresponds to the evaluation of a Boolean function.
I The average path length of a BDD is the length of the paths
from the source node to a terminal node averaged over all
possible inputs. Here, 2+2+3+3+3+3
8 = 2.
Elena Dimitrova Canalyzation in mathematical modeling
62. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Unate cascade functions
I The unate cascade functions (UCFs) make up the unique class
of Boolean functions with the smallest average path length on
a BDD [Mat05].
I This smallest average path length is 2 1
2n1 where n is the
number of variables.
I Thus UCFs can be evaluated more quickly than any other
Boolean function.
I The classes of UCFs and NCFs are identical [JRL07].
Elena Dimitrova Canalyzation in mathematical modeling
64. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Unate cascade functions
I The unate cascade functions (UCFs) make up the unique class
of Boolean functions with the smallest average path length on
a BDD [Mat05].
I This smallest average path length is 2 1
2n1 where n is the
number of variables.
I Thus UCFs can be evaluated more quickly than any other
Boolean function.
I The classes of UCFs and NCFs are identical [JRL07].
Elena Dimitrova Canalyzation in mathematical modeling
66. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
BDD of a UCF/NCF
Recall f (x1; x2; x3) = x2 _ (:x1 ^ x3) is a NCF with b = [1; 0; 0],
a = [1; 1; 0], and = [2; 1; 3]:
Average path length = 3+3+2+1
8 = 1:125.
Elena Dimitrova Canalyzation in mathematical modeling
68. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Number of NCFs
I There is a recursive formula for the number of UCFs.
I Since UCFs are identical to NCFs, the formula can be used for
counting the number of NCFs in n variables, NCF(n):
NCF(n) = 2 E(n), where E(1) = 1; E(2) = 4 and, for n 3,
E(n) =
Xn1
r=2
n
r 1
2r1 E(n r + 1) + 2n:
Elena Dimitrova Canalyzation in mathematical modeling
70. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Other results on NCFs
I NCFs were generalized to an arbitrary
71. eld by D. Murrugarra
[Tom10] who also showed that some features of the dynamics
of these functions is similar to their Boolean counterparts.
I A.S. Jarrah and F. Hinkelmann developed an algorithm that
identi
75. nition
Properties of NCFs
Structure
NCFs and unate cascade functions
Other results
Other results on NCFs
I NCFs were generalized to an arbitrary
76. eld by D. Murrugarra
[Tom10] who also showed that some features of the dynamics
of these functions is similar to their Boolean counterparts.
I A.S. Jarrah and F. Hinkelmann developed an algorithm that
identi
80. nitions
Depth and stability
Derrida plots
Conclusions
I Recall NCFs:
f =
8
:
b1 if x(1) = a1
b2 if x(1)6= a1 and x(2) = a2
b3 if x(1)6= a1 and x(2)6= a2 and x(3) = a3
bn if x(1)6= a1 and and x(n1)6= an1 and x(n) = an
... ...
:bn if x(1)6= a1 and and x(n)6= an
I Quite restrictive!
Elena Dimitrova Canalyzation in mathematical modeling
82. nitions
Depth and stability
Derrida plots
Conclusions
I Recall NCFs:
f =
8
:
b1 if x(1) = a1
b2 if x(1)6= a1 and x(2) = a2
b3 if x(1)6= a1 and x(2)6= a2 and x(3) = a3
bn if x(1)6= a1 and and x(n1)6= an1 and x(n) = an
... ...
:bn if x(1)6= a1 and and x(n)6= an
I Quite restrictive!
Elena Dimitrova Canalyzation in mathematical modeling
84. nitions
Depth and stability
Derrida plots
Conclusions
Nested canalyzing depth
The nested canalyzing depth of a Boolean function f is the largest
index (i) such that the variables x(1); : : : ; x(i) exhibit the nested
canalyzing structure in the de
85. nition of a NCF.
Remarks:
I We make the convention that constant functions are NCFs
and have depth n.
I If a function has i nested canalyzing variables and the
function in the remaining n i variables is constant, this
function also has depth n.
I Functions with no canalyzing variables have depth zero.
Elena Dimitrova Canalyzation in mathematical modeling
87. nitions
Depth and stability
Derrida plots
Conclusions
Example: Nested canalyzing depth
I x2 is canalyzing with canalyzing input 0 and output 0.
I (1) = 2 and b1 = a1 = 0; (2) = 3, b2 = 1, and a2 = 0.
I The remaining function in x1 and x4 is neither canalyzing nor
constant, so f has depth 2.
Elena Dimitrova Canalyzation in mathematical modeling
89. nitions
Depth and stability
Derrida plots
Conclusions
Activity
I Kauman and Shmulevich explored the in
uence of the
function's variables on its output [SK04].
I The partial derivative of a function f (x1; : : : ; xn) with respect
to the variable xj is de
90. ned as @f (x)
@xj
= f (xj ;0) f (xj ;1): Here,
xj ;i = (x1; : : : ; xj1; i ; xj+1; : : : ; xn) and denotes the XOR
function.
I Activity of a variable: The average of the partial derivatives of
j = 1
a variable over all possible inputs, f
2k
P
x2f0;1gk
@f (x)
@xj
:
Elena Dimitrova Canalyzation in mathematical modeling
92. nitions
Depth and stability
Derrida plots
Conclusions
Activity
I Kauman and Shmulevich explored the in
uence of the
function's variables on its output [SK04].
I The partial derivative of a function f (x1; : : : ; xn) with respect
to the variable xj is de
93. ned as @f (x)
@xj
= f (xj ;0) f (xj ;1): Here,
xj ;i = (x1; : : : ; xj1; i ; xj+1; : : : ; xn) and denotes the XOR
function.
I Activity of a variable: The average of the partial derivatives of
j = 1
a variable over all possible inputs, f
2k
P
x2f0;1gk
@f (x)
@xj
:
Elena Dimitrova Canalyzation in mathematical modeling
95. nitions
Depth and stability
Derrida plots
Conclusions
Activity
I Kauman and Shmulevich explored the in
uence of the
function's variables on its output [SK04].
I The partial derivative of a function f (x1; : : : ; xn) with respect
to the variable xj is de
96. ned as @f (x)
@xj
= f (xj ;0) f (xj ;1): Here,
xj ;i = (x1; : : : ; xj1; i ; xj+1; : : : ; xn) and denotes the XOR
function.
I Activity of a variable: The average of the partial derivatives of
j = 1
a variable over all possible inputs, f
2k
P
x2f0;1gk
@f (x)
@xj
:
Elena Dimitrova Canalyzation in mathematical modeling
98. nitions
Depth and stability
Derrida plots
Conclusions
Sensitivity
I The sensitivity of a function quanti
99. es the sensitivity of the
output to variations in the function inputs.
PI It is given by sf (x) =
k
i=1 [f (x ei )6= f (x)] ; where ei
denotes the ith unit vector and is an indicator function.
I Essentially computing the number of inputs with Hamming
distance one from an input x that gives a dierent function
output than x.
I The average sensitivity of a function, sf , is the expected value
of sf (x), sf = E
sf (x)
=
Pk
i=1 f
i :
Elena Dimitrova Canalyzation in mathematical modeling
101. nitions
Depth and stability
Derrida plots
Conclusions
Sensitivity
I The sensitivity of a function quanti
102. es the sensitivity of the
output to variations in the function inputs.
PI It is given by sf (x) =
k
i=1 [f (x ei )6= f (x)] ; where ei
denotes the ith unit vector and is an indicator function.
I Essentially computing the number of inputs with Hamming
distance one from an input x that gives a dierent function
output than x.
I The average sensitivity of a function, sf , is the expected value
of sf (x), sf = E
sf (x)
=
Pk
i=1 f
i :
Elena Dimitrova Canalyzation in mathematical modeling
104. nitions
Depth and stability
Derrida plots
Conclusions
Sensitivity
I The sensitivity of a function quanti
105. es the sensitivity of the
output to variations in the function inputs.
PI It is given by sf (x) =
k
i=1 [f (x ei )6= f (x)] ; where ei
denotes the ith unit vector and is an indicator function.
I Essentially computing the number of inputs with Hamming
distance one from an input x that gives a dierent function
output than x.
I The average sensitivity of a function, sf , is the expected value
of sf (x), sf = E
sf (x)
=
Pk
i=1 f
i :
Elena Dimitrova Canalyzation in mathematical modeling
107. nitions
Depth and stability
Derrida plots
Conclusions
Sensitivity
I The sensitivity of a function quanti
108. es the sensitivity of the
output to variations in the function inputs.
PI It is given by sf (x) =
k
i=1 [f (x ei )6= f (x)] ; where ei
denotes the ith unit vector and is an indicator function.
I Essentially computing the number of inputs with Hamming
distance one from an input x that gives a dierent function
output than x.
I The average sensitivity of a function, sf , is the expected value
of sf (x), sf = E
sf (x)
=
Pk
i=1 f
i :
Elena Dimitrova Canalyzation in mathematical modeling
110. nitions
Depth and stability
Derrida plots
Conclusions
Sensitivity
I Shmulevich and Kauman suggest that networks created using
functions that are less sensitive will be more dynamically
ordered (stable) than those with higher sensitivity.
I They show in [SK04] that a random Boolean function in n
variables has average sensitivity n2
.
I Also, for a Boolean function with depth at least one, the
14
14
expected activities
of the variables (x1; : : : ; xn) are given by ; ; ; : : : ; , and hence the average sensitivity is n+1
.
14
124 Elena Dimitrova Canalyzation in mathematical modeling
112. nitions
Depth and stability
Derrida plots
Conclusions
Sensitivity
I Shmulevich and Kauman suggest that networks created using
functions that are less sensitive will be more dynamically
ordered (stable) than those with higher sensitivity.
I They show in [SK04] that a random Boolean function in n
variables has average sensitivity n2
.
I Also, for a Boolean function with depth at least one, the
14
14
expected activities
of the variables (x1; : : : ; xn) are given by ; ; ; : : : ; , and hence the average sensitivity is n+1
.
14
124 Elena Dimitrova Canalyzation in mathematical modeling
114. nitions
Depth and stability
Derrida plots
Conclusions
Sensitivity
I Shmulevich and Kauman suggest that networks created using
functions that are less sensitive will be more dynamically
ordered (stable) than those with higher sensitivity.
I They show in [SK04] that a random Boolean function in n
variables has average sensitivity n2
.
I Also, for a Boolean function with depth at least one, the
14
14
expected activities
of the variables (x1; : : : ; xn) are given by ; ; ; : : : ; , and hence the average sensitivity is n+1
.
14
124 Elena Dimitrova Canalyzation in mathematical modeling
116. nitions
Depth and stability
Derrida plots
Conclusions
Depth and sensitivity
I When reverse engineering gene networks, a nested canalyzing
structure may apply to some of the variables, while the
behavior of the remaining variables may be ambiguous or
unknown.
I We say that a function has depth at least d if d of the
variables are known to be nested canalyzing, while the
remaining n d variables can take on any Boolean function,
canalyzing or otherwise.
I We'll show that the larger the value of d is, the less sensitive
the function will be on average.
Elena Dimitrova Canalyzation in mathematical modeling
118. nitions
Depth and stability
Derrida plots
Conclusions
Depth and sensitivity
I When reverse engineering gene networks, a nested canalyzing
structure may apply to some of the variables, while the
behavior of the remaining variables may be ambiguous or
unknown.
I We say that a function has depth at least d if d of the
variables are known to be nested canalyzing, while the
remaining n d variables can take on any Boolean function,
canalyzing or otherwise.
I We'll show that the larger the value of d is, the less sensitive
the function will be on average.
Elena Dimitrova Canalyzation in mathematical modeling
120. nitions
Depth and stability
Derrida plots
Conclusions
Depth and sensitivity
I When reverse engineering gene networks, a nested canalyzing
structure may apply to some of the variables, while the
behavior of the remaining variables may be ambiguous or
unknown.
I We say that a function has depth at least d if d of the
variables are known to be nested canalyzing, while the
remaining n d variables can take on any Boolean function,
canalyzing or otherwise.
I We'll show that the larger the value of d is, the less sensitive
the function will be on average.
Elena Dimitrova Canalyzation in mathematical modeling
122. nitions
Depth and stability
Derrida plots
Conclusions
Depth and sensitivity
Extending Kauman and Shmuleviche's result to functions of depth at
least d, we have the following result.
Result (Layne, D., Macauley [LDM12])
Let fd be a Boolean function in n variables with nested canalyzing depth
at least d. Renumbering the variables if necessary, assume that x1; : : : ; xd
are the nested canalyzing variables with (i) = i . Then the expected
activities of the variables (x1; : : : ; xn) are given by
E
fd
=
1
2
;
1
4
; : : : ;
1
2d ;
1
2d+1 ; : : : ;
1
2d+1
:
Further, the average sensitivity of fd is
E[sfd ] =
n d
2d+1 +
Xd
i=1
1
2d =
n d
2d+1 + 1
1
2d :
Elena Dimitrova Canalyzation in mathematical modeling
124. nitions
Depth and stability
Derrida plots
Conclusions
Depth and sensitivity
Now, we see that
E[sfd ] E[sfd+1] = 1
1
2d +
n d
2d+1
1
1
2d+1 +
n d 1
2d+2
=
n d 1
2d+2 ;
which rapidly goes to zero, so each subsequent canalyzing variable
has a much smaller impact on the sensitivity.
I Thus, the dierence in sensitivity between fully nested
canalyzing functions and partially nested canalyzing
functions of sucient depth is very slight.
Elena Dimitrova Canalyzation in mathematical modeling
126. nitions
Depth and stability
Derrida plots
Conclusions
Moral
Using strictly NCFs for modeling when stability is a desirable
property is an overkill. PNCFs of sucient depth are good enough.
Elena Dimitrova Canalyzation in mathematical modeling
128. nitions
Depth and stability
Derrida plots
Conclusions
Derrida plots
x1(t) and x2(t): two states in random Boolean network.
(t) = 1
n jjx1(t) x2(t)jj1 (normalized Hamming distance), where
jj jj1 is the standard `1 metric.
A Derrida curve is a plot of (t + 1) vs. (t) arranged uniformly
over dierent states and networks [DP86].
Elena Dimitrova Canalyzation in mathematical modeling
130. nitions
Depth and stability
Derrida plots
Conclusions
Phases
Consider the curve for small values of (t).
I Frozen phase: Curve lies below y = x. The phase space of
such networks consists of many
131. xed points and small
attractor cycles.
I Chaotic phase: Curve lies above y = x. Small perturbations
propagate throughout the network.
I Critical phase: The boundary threshold between the frozen
and chaotic phases.
There is evidence that many biological and social networks lie in
the critical phase, as they must be stable enough to endure
changes in the environment, yet
exible enough to adapt when
necessary [BABC+08, NPA+08, NPL+08, SKA05].
Elena Dimitrova Canalyzation in mathematical modeling
135. nitions
Depth and stability
Derrida plots
Conclusions
Observations
I Networks of larger depth show more orderly dynamics than
those of smaller depth.
I The curves move closer together as depth increases, i.e.,
marginal bene
136. t of stability as depth increases.
I Observations from Derrida plots match theoretical results.
Elena Dimitrova Canalyzation in mathematical modeling
138. nitions
Depth and stability
Derrida plots
Conclusions
Conclusions
I PNCFs are a reasonable generalization of NCFs.
I Stability increases as depth increases, however the marginal
gain in stability drops o quickly.
I Just a few degrees of canalyzation are necessary to drop the
network into critical regime.
Elena Dimitrova Canalyzation in mathematical modeling
142. nitions
Depth and stability
Derrida plots
Conclusions
[BABC+08] E. Balleza, E. Alvarez-Buylla, A. Chaos, S.A. Kauman, I. Shmulevich, and M. Aldana.
Critical dynamics in genetic regulatory networks: Examples from four kingdoms.
PLoS ONE, 3:e2456, 2008.
[DP86] B. Derrida and Y. Pomeau.
Random networks of automata: a simple annealed approximation.
Europhys. Lett., 1:45{49, 1986.
[JH10] Abdul Salam Jarrah and Franziska Hinkelmann.
Inferring biologically relevant models: Nested canalyzing functions.
In Algebraic and Numeric Biology 2010, 2010.
[JL07] Abdul Salam Jarrah and Reinhard C. Laubenbacher.
Discrete models of biochemical networks: The toric variety of nested canalyzing functions.
In AB, pages 15{22, 2007.
[JRL07] Abdul Salam Jarrah, Blessilda Raposa, and Reinhard Laubenbacher.
Nested canalyzing, unate cascade, and polynomial functions.
Physica D. Nonlinear phenomena, 233:167{174, 2007.
[Kau93] Stuart Kauman.
The Origins of Order: Self-Organization and Selection in Evolution.
Oxford University Press, 1993.
[KPST03] Stuart Kauman, Carsten Peterson, Bjorn Samuelsson, and Carl Troein.
Random boolean network models and the yeast transcriptional network.
PNAS, 100(25):14796{14799, 2003.
[KPST04] Stuart Kauman, Carsten Peterson, Bjorn Samuelsson, and Carl Troein.
Genetic networks with canalyzing boolean rules are always stable.
Proceedings of the National Academy of Sciences of the United States of America,
101(49):17102{17107, December 2004.
Elena Dimitrova Canalyzation in mathematical modeling
144. nitions
Depth and stability
Derrida plots
Conclusions
[LDM12] L. Layne, E.S. Dimitrova, and M. Macauley.
Nested canalyzing depth and network stability.
Bulletin of Mathematical Biology, 74:422{433, 2012.
[Lyn95] J. F. Lynch.
On the threshold of chaos in random boolean cellular automata.
Random Struct. Algorithms, 6(2{3):239{260, 1995.
[Mat05] Munehiro Matsuura.
Average path length of binary decision diagrams.
IEEE Trans. Comput., 54(9):1041{1053, 2005.
Fellow-Butler, Jon T. and Fellow-Sasao, Tsutomu.
[NFW07] S. Nikolajewa, M. Friedel, and T. Wilhelm.
Boolean networks with biologically relevant rules show ordered behavior.
BioSystems, 90:40{47, 2007.
[NPA+08] M. Nykter, N.D. Price, M. Aldana, S.A. Ramsey, S.A. Kauman, L.E. Hood, O. Yli-Harja, and
I. Shmulevich.
Gene expression dynamics in the macrophage exhibit criticality.
Proc. Natl. Acad. Sci., 150:1897{1900, 2008.
[NPL+08] M. Nykter, N.D. Price, A. Larjo, T. Aho, S.A. Kauman, O. Yli-Harja, and I. Shmulevich.
Critical networks exhibit maximal information diversity in structure-dynamics relationships.
Phys. Rev. Lett., 100:058702, 2008.
[SK04] I. Shmulevich and S.A. Kauman.
Activities and sensitivities in boolean network models.
Phys Rev Lett, 93(4):048701, 2004.
Elena Dimitrova Canalyzation in mathematical modeling
146. nitions
Depth and stability
Derrida plots
Conclusions
[SKA05] I. Shmulevich, S.A. Kauman, and M. Aldana.
Eukaryotic cells are dynamically ordered or critical but not chaotic.
Proc. Natl. Acad. Sci., 102:13439{13444, 2005.
[Sta87] D. Stauer.
On forcing functions in kaumans random boolean networks.
Journal of Statistical Physics, 46(3{4):789794, 1987.
[Ste99] M. D. Stern.
Emergence of homeostasis and noise imprinting in an evolution model.
PNAS, 96(19):10746{10751, 1999.
[Tom10] D. M. Tomairo.
Multi-states nested canlyzing functions.
Preprint, 2010.
[Wad42] C. H. Waddington.
Canalisation of development and the inheritance of acquired characters.
Nature, 150:563{564, 1942.
Elena Dimitrova Canalyzation in mathematical modeling