Part 2B: Pattern Formation
2/2/16
Differentiation
& Pattern Formation
• A central problem in
development: How do cells
differentiate to fulfill
different purposes?
• How do complex systems
generate spatial & temporal
structure?
• CAs are natural models of
intercellular communication
B.
Pattern Formation
2/2/16
1
2/2/16
2
Vermiculated Rabbit Fish
Plecostomus
2/2/16
photos ©2000, S. Cazamine
3
2/2/16
Zebra
figs . from Camazine & al.: Self-Or g. Biol. Sys .
4
Activation & Inhibition
in Pattern Formation
• Color patterns typically have a characteristic length scale
• Independent of cell size and animal size
• Achieved by:
– short-range activation ⇒ local uniformity
– long-range inhibition ⇒ separation
2/2/16
figs . from Camazine & al.: Self-Or g. Biol. Sys .
5
2/2/16
6
1
Part 2B: Pattern Formation
2/2/16
Interaction Parameters
CA Activation/Inhibition Model
•
•
•
•
$
'
si ( t + 1) = sign&h + J1 ∑ s j ( t ) + J 2 ∑ s j ( t ))
&%
)(
rij <R1
R1 ≤rij <R 2
• R1 and R2 are the interaction ranges
• J1 and J2 are the interaction strengths
2/2/16
Let states si ∈ {–1, +1}
and h be a bias parameter
and rij be the distance between cells i and j
Then the state update rule is:
7
2/2/16
8
€
Example
(R1 =1, R2 =6, J1 =1, J2 =–0.1, h=0)
Demonstration of NetLogo
Program for Activation/Inhibition
Pattern Formation
RunAICA.nlogo
2/2/16
9
fig s. fro m Bar-Yam
2/2/16
Effect of Bias
Effect of Interaction Ranges
(h = –6, –3, –1; 1, 3, 6)
R2 = 6
R1 = 1
h=0
R2 = 8
R1 = 1
h=0
R2 = 6
R 1 = 1.5
h=0
2/2/16
fig s. fro m Bar-Yam
10
11
2/2/16
R2 = 6
R 1 = 1.5
h = –3
fig s. fro m Bar-Yam
12
2
Part 2B: Pattern Formation
2/2/16
Digression on Diffusion
Differential Interaction Ranges
• Simple 2-D diffusion equation:
• How can a system using strictly local
interactions discriminate between states at
long and short range?
• E.g. cells in developing organism
• Can use two different morphogens diffusing
at two different rates
13
0 $! ∇ 2 A
&#
DI &%#" ∇ 2 I
$ ! fA ( A, I )
&+#
& # f A, I
)
% " I(
∂ 2 A( x, y ) ∂ 2 A( x, y )
+
∂x 2
∂y 2
– positive in a local minimum
– negative in a local maximum
2/2/16
14
General Reaction-Diffusion System
Reaction-Diffusion System
!
∂ ! A $ # DA
#
&=
∂ t " I % #" 0
∇ 2 A( x, y ) =
€
2/2/16
diffusion
• Recall the 2-D Laplacian:
• The Laplacian (like 2nd derivative) is:
– activator diffuses slowly (short range)
– inhibitor diffuses rapidly (long range)
∂A
= DA ∇ 2 A + fA ( A, I )
∂t
∂I
= DI ∇ 2 I + fI ( A, I )
∂t
A ( x, y ) = D∇ 2 A ( x, y )
# n
&
∂ci
= ∑ kαν iα % ∏ ckmkα ( − ∇ ⋅ ji
∂t
$ k=1
'
α

where ji = µi ci − div Di ci (flux)
reaction
where kα = rate constant for reaction α
$
&
&
%
and ν iα = stoichiometric coefficient
and mkα = a non-negative integer

and µi = drift vector
# A&
c˙ = D∇ 2c + f (c), where c = % (
$I'
2/2/16
and Di = diffusivity matrix
where div Dc = ∑ e j ∑ D jk ∂c
15
2/2/16
j
k
∂xk
16
€
Framework for Complexity
NetLogo Simulation of
Reaction-Diffusion System
• change = source terms + transport terms
• source terms = local coupling
= interactions local to a small region
• transport terms = spatial coupling
= interactions with contiguous regions
= advection + diffusion
1. Diffuse activator in X and Y directions
2. Diffuse inhibitor in X and Y directions
3. Each patch performs:
stimulation = bias + activator – inhibitor + noise
if stimulation > 0 then
set activator and inhibitor to 100
else
set activator and inhibitor to 0
– advection: non-dissipative, time-reversible
– diffusion: dissipative, irreversible
2/2/16
17
2/2/16
18
3
Part 2B: Pattern Formation
2/2/16
Continuous-time Activator-Inhibitor System
• Activator A and inhibitor I may diffuse at different
rates in x and y directions
• Cell becomes more active if activator + bias
exceeds inhibitor
• Otherwise, less active
• A and I are limited to [0, 100]
(depletion/saturation)
Demonstration of NetLogo
Program for Activator/Inhibitor
Pattern Formation
∂A
∂2A
∂2A
= dAx 2 + dAy 2 + k A ( A + B − I )
∂t
∂x
∂y
Run Pattern.nlogo
∂I
∂ 2I
∂ 2I
= dIx 2 + dIy 2 + k I ( A + B − I )
∂t
∂x
∂y
2/2/16
19
• Alan Turing studied the mathematics of
reaction-diffusion systems
• Turing, A. (1952). The chemical basis of
morphogenesis. Philosophical Transactions
of the Royal Society B 237: 37–72.
Run Activator-Inhibitor.nlogo
• The resulting patterns are known as Turing
patterns
21
2/2/16
22
A Key Element of
Self-Organization
Observations
• Activation vs. Inhibition
•
•
•
•
With local activation and lateral inhibition
And with a random initial state
You can expect to get Turing patterns
These are stationary states (dynamic
equilibria)
• Macroscopically, Class I behavior
• Cooperation vs. Competition
• Amplification vs. Stabilization
• Growth vs. Limit
• Positive Feedback vs. Negative Feedback
– Positive feedback creates
– Microscopically, may be class III
2/2/16
20
Turing Patterns
Demonstration of NetLogo
Program for Activator/Inhibitor
Pattern Formation
with Continuous State Change
2/2/16
2/2/16
– Negative feedback shapes
23
2/2/16
24
4
Part 2B: Pattern Formation
2/2/16
Image Processing in BZ Medium
Reaction-Diffusion Computing
• Has been used for image processing
– diffusion ⇒ noise filtering
– reaction ⇒ contrast enhancement
• Depending on parameters, RD computing
can:
– restore broken contours
– detect edges
– improve contrast
2/2/16
• (A) boundary detection, (B) contour enhancement,
(C) shape enhancement, (D) feature enhancement
25
Voronoi Diagrams
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
26
27
• Collision-free path planning
• Determination of service areas for power
substations
• Nearest-neighbor pattern classification
• Determination of largest empty figure
2/2/16
Computation of Voronoi Diagram
by Reaction-Diffusion Processor
2/2/16
Imag e < Ad amatzk y , Co mp . in No nlin ea r Med ia & Auto m. Co ll.
Some Uses of Voronoi Diagrams
• Given a set of generating
points:
• Construct a polygon
around each generating
point of set, so all points
in a polygon are closer to
its generating point than to
any other generating
points.
2/2/16
2/2/16
29
28
Mixed Cell Voronoi Diagram
2/2/16
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
30
5
Part 2B: Pattern Formation
2/2/16
Path Planning via BZ medium:
No Obstacles
2/2/16
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
Path Planning via BZ medium:
Circular Obstacles
31
2/2/16
Mobile Robot with Onboard
Chemical Reactor
2/2/16
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
32
Actual Path: Pd Processor
33
2/2/16
Actual Path: Pd Processor
2/2/16
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
34
Actual Path: BZ Processor
35
2/2/16
Imag e < Ad amatzk y & al., Rea ctio n -Diffu sio n Co mpu ters
36
6
Part 2B: Pattern Formation
2/2/16
Bibliography for
Reaction-Diffusion Computing
1. Adamatzky, Adam. Computing in Nonlinear
Media and Automata Collectives. Bristol: Inst.
of Physics Publ., 2001.
2. Adamatzky, Adam, De Lacy Costello, Ben, &
Asai, Tetsuya. Reaction Diffusion Computers.
Amsterdam: Elsevier, 2005.
2/2/16
37
7