Content uploaded by Bradly Alicea
Author content
All content in this area was uploaded by Bradly Alicea on Dec 20, 2016
Content may be subject to copyright.
Morphozoic, Cellular Automata with Nested
Neighborhoods as a Metamorphic
Representation of Morphogenesis
Thomas Portegys
Dialectek, USA
Gabriel Pascualy
University of Michigan, USA
Richard Gordon
Gulf Specimen Marine Laboratory & Aquarium, USA & Wayne State University, USA
Stephen P. McGrew
New Light Industries, USA
Bradly Alicea
Orthogonal Research, Champaign, IL, USA
Abstract
A cellular automaton model, Morphozoic, is presented. Morphozoic may be used to investigate the
computational power of morphogenetic fields to foster the development of structures and cell
differentiation. The term morphogenetic field is used here to describe a generalized abstraction: a cell
signals information about its state to its environment and is able to sense and act on signals from nested
neighborhood of cells that can represent local to global morphogenetic effects. Neighborhood signals
are compacted into aggregated quantities, capping the amount of information exchanged: signals from
smaller, more local neighborhoods are thus more finely discriminated, while those from larger, more
global neighborhoods are less so. An assembly of cells can thus cooperate to generate spatial and
temporal structure. Morphozoic was found to be robust and noise tolerant. Applications of Morphozoic
presented here include: (1) Conway’s Game of Life, (2) cell regeneration, (3) evolution of a gastrulation-
like sequence, (4) neuron pathfinding, and (5) Turing’s reaction-diffusion morphogenesis.
Keywords: Morphogenesis, Cellular automata, Moore neighborhoods
Introduction
Morphogenesis is a biological process by which cells move and differentiate into organs and tissues
through genetic expression and collaborative, often physical mechanisms. They become different kinds
of cells, perhaps as many as 7000 kinds in our bodies (Gordon, 1999). One of the most persistent
concepts of morphogenesis is the morphogenetic field (Beloussov, Opitz & Gilbert, 1997; Alberts et al.,
2002; Levin, 2011, 2012; Morozova & Shubin, 2012; Vecchi & Hernández, 2014; Beloussov, 2015) with
clinical significance for human birth defects (Opitz & Neri, 2013). A morphogenetic field is a region of an
embryo that has the potential to develop into a specific structure. How this happens has been subject to
much investigation and debate. Some mechanisms are better understood than others. As discussed
in (Tyler, 2014), which reviews the full panoply of models of morphogenetic fields, we have added the
idea that a morphogenetic field is the trajectory of a two dimensional differentiation wave that triggers
a step of differentiation in each cell it traverses (Gordon, 1999; Gordon & Gordon, 2016b, a).
While the mechanisms behind the transformation of an egg into an embryo (embryogenesis) and then
to an adult organism have traditionally been of great interest to biologists, as a pattern formation
problem it is equally intriguing to computer scientists. Part of the allure involves the spontaneous
attainment of order of great complexity from geometrically simple beginnings. Even though one of the
founding fathers of computer science (Leavitt, 2006; Hodges, 2014; Tyldum, 2014), Alan Turing
proposed a plausible model for understanding one level of the self-organizing aspect of
morphogenesis (Turing, 1952; Gordon, 2015), the process of phenotype-building has a “ghost in the
machine” (Koestler, 1967), a cybernetic aspect (Gordon & Gordon, 2016a; Gordon & Stone, 2016) that
has gone underappreciated.
Recent chemical experiments have revealed that while Turing’s original reaction-diffusion equations
portray certain aspects of morphogenesis, they do not account for heterogeneity (Tompkins et al., 2014)
or the multistep hierarchical differentiation of cells into different types (Gordon, 2015). In this study, we
propose that given the right representation, simulated morphogenesis can yield solutions that are
biologically plausible. Our approach, Morphozoic, models a hierarchical structure of cellular
communities. Computationally, these communities are nested versions of Moore-like neighborhoods. A
Moore neighborhood is the set of cells that are immediate neighbors to a cell, so in a two-dimensional
square array the Moore neighborhoods contain eight cells (Weisstein, 2016b). In Morphozoic, a single
higher-level cell houses an entire lower-level Moore neighborhood, down to single cells, and a set of
lower-level neighborhoods compose a higher-level neighborhood (Gordon & Rangayyan, 1984a, b).
While this serves as a constraint on cell-cell communication, it also serves as top-down information. This
top-down information, when coupled with local, bottom-up information at different spatial scales,
provides us with a mechanism for strongly emergent phenomena (Holland, 1992).
We originally called such nested neighborhoods “adaptive neighborhoods” in the context of image
processing, because around each pixel a Moore neighborhood size was chosen that best fit the local
features of the picture (Gordon & Rangayyan, 1984a) (Gordon & Rangayyan, 1984b). The field of
adaptive neighborhood image processing now has over 400 publications, and an independent discovery
of the idea has extended the idea in many additional directions (Katkovnik, Egiazarian & Astola, 2006). In
the field of cellular automata, adaptive neighborhoods have been used in the sense of changing
neighborhood type rather than size (Mofrad et al., 2015). Our approach of nested neighborhoods has
been combined with cellular automata for edge detection (Liu et al., 2012) and rule
identification (Adamatzky, 1997; Sun, Rosin & Martin, 2011; Zhao, Wei & Billings, 2012). Irregular (Batty,
2003) and “extended” (Guan & Clarke, 2010) Moore neighborhoods have also been used for geographic
cellular automata, though not with the nesting idea in mind. Irregular, grown adaptive neighborhoods
have been called coalitions in cellular automata (Burguillo, 2013). Morphozoic appears to be unique in
using nested neighborhoods, not to find an optimal neighborhood of a cell, but to provide information
at many scales to that cell. It thus permits the study of local/global interactions.
Computer modeling of biological systems is widespread (Wyczalkowski et al., 2012; Tanaka, 2015). The
Morphozoic approach is based on the Cellular Automaton (CA) architecture which exhibits
computational universality (Dobrescu & Purcarea, 2011) (Wolfram, 2001) that is not well understood in
the context of biological development. The aim of the Morphozoic project is to build an abstract model
of morphogenetic fields to explore its computational capabilities. While the model may lead to some
insights into biology, this is not a central goal of the project. Simulations presented here suggest that
the model can be used to produce general self-organizing structures. In particular, many aspects of
modern human life involve local/global interactions, so Morphozoic may contribute to the social
sciences (Batten, 2001). We also show how Morphozoic may be used to reverse engineer a sequence of
state changes of a system and derive an approximation to the rules governing that system. Morphozoic
may therefore be used for reverse engineering (Gordon & Melvin, 2003; Deutsch, Maini & Dormann,
2007; Elmenreich & Fehervari, 2011; Lobo & Levin, 2015).
Because of its local/global construction, Morphozoic may be a step towards meeting the challenge
posed by Russ Abbott:
“...when a glider appears in the Game of Life, it has no effect on the how the system behaves. The agents don’t see a glider
coming and duck. More significantly we don’t know how to build systems so that agents will be able to notice gliders and duck.
It would be an extraordinary achievement in artificial intelligence to build a modeling system that could notice emergent
phenomena and see how they could be exploited. Yet we as human beings do this all the time” (Abbott, 2006).
The Morphozoic platform can model a wide range of natural and artificial phenomena, but the question
remains whether or not the software can exhibit biological realism. Certainly in terms of approximating
morphogenesis, it is not clear whether patterns formed and identified by Morphozoic are produced by
biologically-realistic mechanisms. However, in the realm of biological realism, Morphozoic is consistent
with similar approximations of biological complexity. Particularly at the level of interacting cells,
morphogenesis and Morphozoic alike exhibit what Abbott (Abbott, 2006) calls ‘epiphenomena’. These
epiphenomena, or emergent outcomes of collective interactions between agents, are as biologically
realistic at the macro-scale as gene action is at the micro-scale. However, as morphogenesis is non-
reductive (Abbott, 2006; Gordon & Gordon, 2016a), it becomes difficult to make exact predictions of
behavior in existing biological systems. What makes for an excellent naturalistic pattern replication
mechanism (Wolfram, 1984) might make for a poor descriptor of unfolding processes in the frog
embryo.
Biological realism in modeling involves the degree to which selectively adding in features of the system
you are attempting to model produces a useful representation. At a macro-level of description,
Morphozoic exhibits a priori biological realism (Bourgine & Lesne, 2010) that captures the higher-order
dynamics of a biological system rather than the lower-level causal mechanisms of complexity (Ruxton &
Saravia, 1998). A priori realism incorporates known features in a minimal and abstract fashion (Bourgine
& Lesne, 2010). In cases where the underlying system has many moving parts and layers of complexity, a
high degree of abstraction is required to avoid transcomputational limits (Bremermann, 1967; Ashby,
1968; Gordon, 1970; Bower, 2005). Transcomputational limits are of particular concern in biological
systems, where the myriad sources of variation can produce a very high dimensional problem space.
While abstraction in the face of transcomputational limits is a supposed requirement for biologically-
inspired simulations, nevertheless many models of collective behavior (Resnick, 1994) and
evolution (Adami, 1998; Alicea & Gordon, 2014) also utilize highly abstract representation while
producing realistic biological system dynamics. This is one reason why Morphozoic can reproduce
patterns seen in morphogenesis without invoking mechanisms of gene expression.
Another reason why Morphozoic can exhibit “lifelike” patterns is due to the nature of Cellular Automata.
In many cases what drives changes in the dynamics of the model are not lower-level control
mechanisms, but the order in which key spatial and temporal events play out. While this is important for
real biological systems as well, changes to spatial and temporal order of how events are executed lead
to synergistic effects in Cellular Automata dynamics (Ruxton, 1996). Furthermore, the timing of events in
a Cellular Automata model impact predictive ability, as Huberman and Glance (Huberman & Glance,
1993) have shown by implementing asynchronous behaviors in the Prisoner's Dilemma game. In
Morphozoic, we keep the standard features of cellular automata: discrete time steps, each cell changes
state simultaneously with all the others, and no cell moves from its initial position. These three aspects
are clearly not biological. Lifting these constraints is a topic for future research. Morphozoic provides a
rich starting point.
There are also parallels between biological and sociotechnical systems that demonstrate how adaptive
behavior might be as much a consequence of temporal evolution than formal biological mechanisms.
Sometimes this temporal evolution is intertwined with structural features of the system, such as when
scientists can “unboil” an egg and return proteins to their original conformational state (Bijelic et al.,
2015). The origin of evolutionary novelties also relies upon how temporal sequences interact with
existing variation to produce innovations. This, evolutionary novelties are generated from what is
already available through recombination and repurposing (Jacob, 1977). Assembling what already exists
into new combinations is called combinatorial innovation, part of something Wagner and
Rosen (Wagner & Rosen, 2014) refer to as innovability. Acting as a mechanism for new phenotypic
possibilities, characterizing the innovability of emergent systems like Morphozoic might provide a very
broad window into the systems-level mechanisms of developmental processes.
Innovability is a key driver in the functional diversity seen in Morphozoic, and can be analogous to the
types of developmental plasticity observed in biological systems (West-Eberhard, 2002). In phenomena
ranging from axonogenesis to generalized stress response (Bateson et al., 2004; Gluckman, Hanson &
Low, 2011; Low, Gluckman & Hanson, 2012), organisms can exhibit a morphogenetic adaptive response
to both local and global signals. In this sense, Morphozoic provides a means to explore how the timing
and phenotypic consequences of these processes might unfold (Moczek et al., 2011). While Morphozoic
does not possess the same mechanisms that drive biological plasticity, phenotypic changes can be
approximated using discrete computation. While Morphozoic may be able to approximate the pattern-
formation aspects of developmental plasticity, a concept known as the adjacent possible may be used to
bridge the gap between biology and simulation. The adjacent possible was originally proposed by Stuart
Kauffman as a way to describe possible new states for a system given historical contingency (Johnson,
2010). The adjacent possible is ever-expanding, and this non-random expansion into possibility space is
what drives subsequent innovations (Tria et al., 2014).
Fortunately, the cellular automata representation is well-suited to producing novelties that result from
interactivity. Identifying the specific mechanisms for the generation of novelties is a relatively
unexplored question (Kier, Bonchev & Buck, 2005). How can we discover these pathways and make
parallels with developmental and other biological systems? On the one hand, the adjacent possible can
be made salient using a method called fitness optimization. This allows us to understand patterns of
innovation that conform to the adjacent possible idea as a series of moves towards a fitness peak (or
functional optimum). On the other hand, many biological and technological systems also include
significant constraints on their evolution such that the states making up the so-called adjacent possible
are limited to relatively small portions of the whole system (Solé et al., 2013). One way to investigate
these non-innovable portions of the system is to construct a neutral network (van Nimwegen,
Crutchfield & Huynen, 1999) that defines all possible configurations of the system. To approximate the
full set of possible states (which may or may not be transcomputational), we may rely upon
computational information such as the neutral network, cellular automata rules, and knowledge about
the constraints a given system might pose. Applying the concept of the adjacent possible then allows us
to limit the range of plausible emergent mechanisms in biological and non-biological settings.
Coupling Morphozoic with other computational and data-driven models of embryogenesis may lead us
to a middle-out approach (Noble, 2002). The middle-out approach can be defined as a combination of
top-down and bottom-up approaches where specification begins at the level where the data are
sufficient. Using inferential and other methods, we can then move towards other levels of analysis. This
is ideal for systems that are too interactive for systems or reductionist approaches to be effective on
their own (Noble, 2002). The middle-out approach is particularly useful to data-driven research in that it
strikes a balance between reductionism and integration. As the reductionists have mechanisms, the
systems people have overarching descriptions (Kohl & Noble, 2009). Morphozoic occupies a middle-
ground between reductionism and integration in the sense that the states of single cells are influenced
by local neighborhoods, but that each local neighborhood is interlinked to form a potentially large
(global) problem space. This interlinking occurs at many levels, due to the use of nested neighborhoods.
By occupying this middle ground, defined by neighborhoods, we can approach multiscalar biological
processes in novel ways (Walker & Southgate, 2009). Multiscalarity can reach 8 orders of magnitude, as
in diatoms, for instance (Ghobara et al., 2016).
One of the inspirations for Morphozoic is the Morphone model (Portegys, 2002), a programmable,
modular signaling system for morphing complex patterns. One of the ways Morphozoic differs from
Morphone is in its leveraging of local vs. global signaling fields as computing mechanisms. Another
feature of Morphozoic, typical of computing systems but distinct from biological systems, is the use of
cell states as signals to other cells instead of supporting separate signaling objects.
Description
Cellular automaton
Morphozoic is built upon a two-dimensional (2D) cellular automaton (CA) architecture. CAs generally
have these properties:
● A cell has a state value (e.g. on or off). However, the value can be a more complex entity, such as a
real value or vector.
● A cell senses the states of adjacent cells that makes up its neighborhood. One example is a 3x3
Moore neighborhood.
● Action rules determine how a cell state changes based on the states of the cells in its neighborhood.
CAs date back to the 1940s and the work of Stanislaw Ulam, who, requiring a model to study the growth
of crystals, developed a simple lattice network (Ulam, 1962). At the same time, John von Neumann,
Ulam's colleague, was working on the problem of self-replicating systems (von Neumann & Burks, 1966).
Von Neumann's initial design was to have robots build new robots out of a “sea of parts”. This proved
problematic, and a more abstract and discrete model was later developed that became a foundation for
the CA approach. It was Ulam who suggested using a discrete system for creating a reductionist model
of self-replication. Von Neumann’s work in self-replication system is similar to what is probably the most
famous cellular automaton: the “Game of Life,” (Gardner, 1970) which is presented as an application of
Morphozoic in a later section. We introduced long-range interactions in cellular automata in a simple
model of two dimensional “snail” morphogenesis, showing they led to more robust pattern
formation (Gordon, 1966). Long-range effects are also incorporated into human evacuation CA
models (Kaji & Inohara, 2014).
In 2001, Stephen Wolfram’s book A New Kind of Science was published (Wolfram, 2001). The book
discusses how CAs are relevant to the study of biology, chemistry, physics, and all branches of science. In
addition, CAs are relatively easy to create in software and are straightforward in performance
evaluation. For these reasons the CA architecture was chosen as a platform for Morphozoic. As
demonstrated in Wolfram’s work, CAs can serve as a model of parallel distributed processing that fit
many natural systems well. Some of these properties of the physical world:
● Contain a large number of simple parts, called cells, each of which is an automaton. This means
it acts on its own (autonomously).
● Parallel operation.
● Global, emergent effects from local interactions.
Each cell (automaton) can respond to external signals in “deciding” what to do. Those decisions are
restricted to certain choices, either finite in number or represented by continuous values. For example,
CAs can simulate fluid or gas dynamics by storing individual molecules in the cells and implementing
particle interactions by the local rules.
One of the main variations of CAs are in the number of dimensions they contain. A one-dimensional (1D)
CA is simply a row of cells. A two-dimensional (2D) CA is a grid of cells. Three-dimensional (3D) CAs
operate within a volume of discrete cells. The concept of a neighborhood is central to the description of
a CA, as the states of a cell’s neighborhood define the input to the local state transition rules. In a one-
dimensional CA, a cell’s neighborhood is usually its adjacent cells. In a 2D CA, typical neighborhood
configurations are shown in Figure 1. It is also common, as is done for Morphozoic, to include the center
cell in the definition of its neighborhood.
Figure 1– Typical two-dimensional CA neighborhoods. Adapted from Figure 2, Espinola et.al, 2010.
CA cells contain a state that can take on values defined by the automaton. For example, a binary state
can be considered to be in a 0 (off)/1 (on) state, as shown in Figure 2.
Figure 2 – A 2-dimensional grid of cells, each in a discrete state of “0” or “1”. Moore neighborhood of
cells shown inside red bounding box. Automaton for Moore neighborhood is denoted with a red circle.
State values change based on transition rules. A good way to understand how state transition rules work
is to consider a one-dimensional CA with adjacent cell neighborhood, as shown in Figure 3. Using
Wolfram’s terminology, the eight possible neighborhood state configurations and center cell state
transitions can be enumerated. So this rule set is number 30.
Figure 3 – State transition rule set 30, where 30 base 10 = 00011110 base 2. From (Weisstein, 2016a)
with permission per Eric Wolfram's Notice of Copyright http://wolfram.org/copyright.html.
Figure 4 demonstrates how a CA develops over time from a single “on” cell using rule 30. Time advances
from top to bottom.
Figure 4 – Application of rule 30. A single cell is turned on (top row) and in the next time step (second
row) all of the cells are subjected to rule 30. Two of them change state as a result, leaving three cells
turned on in the second row. This process is iterated to generate the third row, etc. Patterns like these
are actually generated by the pigment depositing line of cells in marine cone snails (Waddington &
Cowe, 1969). The discrete version of such one dimensionally generated patterns became known as
Lindenmeyer patterns (de Koster & Lindenmayer, 1987). From (Weisstein, 2016a) with permission per
Eric Wolfram's Notice of Copyright http://wolfram.org/copyright.html
There are many other possible rules to compute a cell’s state from a group of cells. Consider blurring an
image. A pixel’s new state (i.e., its color) is the average of all of its neighbors’ colors. Most image
processing algorithms can be formulated as CAs. The rules define the functionality of the CA.
Variations of CAs
The following are some variations of the CA model:
1. Non-rectangular grids. There is no essential reason why a CA must be confined to a rectangular
grid or space.
2. Probabilistic. The rules of a CA need not necessarily work in a deterministic fashion (Gordon,
1966, 1980). An example is the Stochastic Game of Life (Monetti & Albano, 1997).
3. Continuous. The state of a cell can be something other than a discrete value, such as 0 or 1. For
example, the values could range between 0 and 1. Of course the rules must then reflect how to
calculate these continuous state values. Examples in materials science, traffic flow and
earthquake analysis abound (Olami, Feder & Christensen, 1992; Bubak & Czerwiński, 1999;
Kitakawa, 2004, 2005; Gosálvez et al., 2009; Ferrando et al., 2011; Ferrando, Gosálvez & Colóm,
2012; Li et al., 2015).
4. Historical. In the Game of Life CA, the current state configuration determines the next
configuration. However, taking into account historical states is also possible. For example,
Portegys and Wiles (Portegys & Wiles, 2004) describe a CA that self-repairs in the presence of
noise by use of historical cell states.
5. Moving cells. In these examples, cells have a fixed position on a grid, but can move to other grid
points (or, equivalently, their states can be transferred to other cells or switched with
them) (Gordon et al., 1972, 1975; Chopard, 1990; Fukui & Ishibashi, 1996; Hochberger,
Hoffmann & Waldschmidt, 1999; Halbach & Hoffmann, 2005; Moussa, 2005).
6. Nesting. Another feature of complex systems is that they can be nested into hierarchies. For
example, a city is a complex system of people, a person is a complex system of organs, an organ
is a complex system of cells, etc. Cell values that reflect this hierarchical arrangement are
possible (Weimar, 2001; Dunn & Majer, 2007; Kiester & Sahr, 2008; Dunn, 2010).
From a more speculative viewpoint (Ilachinski, 2001), researchers have raised the question of whether
the universe is a cellular automaton. For example, mathematical models have shown the emergence of
"particles" within CAs, such as the gliders in the Game of Life. This leads to conjectures that the natural
world, which is well described by physics with particle-like objects, could actually be a CA. This
hypothesis has led scholars to a perspective of nature existing within a discrete framework. Edward
Fredkin (Fredkin, 1992), a strong proponent, has proposed the "finite nature hypothesis", i.e., the idea
that "ultimately every quantity of physics, including space and time, will turn out to be discrete and
finite."
Objectives
A main objective is to devise an abstraction that models morphogenesis in a CA using a nested
neighborhood approach. As a type of multilayered scheme (Bandini & Mauri, 1999; Dascalu et al., 2011),
nested neighborhoods, defined more formally below, are simply neighborhoods contained within
neighborhoods, much like Russian matryoshka dolls. Nested neighborhoods provide a straightforward
representation of a morphogenetic field that contains a hierarchy of local vs. global information.
Information about a more local cell neighborhood having fewer cells is more precise and finer grained
than information about a larger, more global neighborhood.
The scheme must be computationally plausible. Neighborhoods of increasing size contain increasing
numbers of cells. In order to constrain the potential information explosion, a specific number of bits are
used to represent each neighborhood, regardless of its size. The smallest neighborhood is represented
precisely; larger neighborhoods are increasingly “fuzzy” because they cannot be completely represented
by the available bits. This forms a precision gradient that decreases as the neighborhood grows in size.
An intended result of this plan is that more distant cells are sensed in aggregation, as described in the
next section.
Additional objectives:
● Compact state change rules.
● Noise tolerant.
● Generalizable from exemplars.
● Evolvable.
Morphogenetic field specification
The morphogenetic field is specified in a CA by equipping cells with these properties:
● A cell state is its type.
● A cell emits, senses, and reacts to signals.
● Signals carry information about the types of neighborhood cells.
● A field is the confluence of signals sensed by each cell.
Rules are embodied in metamorphs, which encapsulate pattern-matching morphogens and cell state
change actions.
Morphogen
A “morphogen” abstracts many types of morphogenesis mechanisms: chemical (the classical definition),
physical, energy, etc. It also summarizes a morphogenetic field as a set of a cell’s nested neighborhoods
and their contents. This is shown in Figure 5. A neighborhood consists of an NxN set of sectors
surrounding a lower level neighborhood:
neighborhoodi =NxN(neighborhoodi-1)
where N is a fixed odd positive number chosen by the user of Morphozoic, and neighborhood0 is
composed of NxN elementary cell sectors.
Hence the number of cells in neighborhoodi = NixNi= N2i.
A morphogen is composed of a set of nested neighborhoods:
morphogen(cell) = { neighborhood0(cell), neighborhood1(neighborhood0), ...
neighborhoodn(neighborhoodn-1) }
The value of a sector is a vector representing a histogram of the cell type densities contained within it:
value(sector) = [ density(cell-type0), density(cell-type1), … density(cell-typen) ]
The number of cells contributing to the density histogram of a sector of neighborhoodi = Ni-1x Ni-1
Figure 5 – Morphogen nested neighborhoods.
Metamorph
A metamorph represents a cellular automaton morphogen→action agent, defined as a mapping from a
morphogen to a cell type.
Generation
A set of metamorphs describing a pattern of cell activity can be generated from manual input or a
programmed sequence of cellular automaton transitions. For example, the Game of Life application uses
the programmed Game of Life rules to process the cell states. As the CA changes, the neighborhoods for
each cell are used to construct morphogens, and the cell type transitions associated with the
morphogens are actions. The morphogens and actions are composed into metamorphs.
Execution
Once generated, the metamorphs can be independently used to “execute” the application. Metamorph
execution consists of creating a morphogen for each cell in the grid and comparing each of these
morphogens to the stored set of morphogens contained in the generated metamorphs, where the
distance between them is given by:
The metamorph having the least morphogen distance is chosen as the cell action.
A unique feature afforded by the use of a distance metric to match morphogens is a noise-tolerant, self-
healing capability. Metamorphs act on cells according to neighborhood similarity, which steers cell
states toward patterns stored in the metamorphs. This feature is a hallmark of biological systems, and is
quite distinct from typical rigid CA rule formulations, such as the Game of Life (see applications), where
a minor introduction of noise often results in global disruptions.
Artificial neural network implementation
A compact, fast, and noise tolerant representation of metamorphs can be implemented by an artificial
neural network (ANN), a biologically-based learning machine that is particularly adept at classifying
input patterns into output categories (Haykin, 2011).
ANN background
ANNs are loosely modeled after the neuronal structure of biological nervous systems but on smaller
scales. A large ANN might have thousands of processor units and interconnections, whereas a human
brain, for example, can have 86 billion neurons. An African elephant brain contains 267 billion
neurons (Wikipedia, 2016b), and human cerebral cortex contains 150 trillion synaptic
interconnections (Drachman, 2005). ANN architectures often functionally diverge from their biological
counterparts in important ways, including how learning is implemented.
An ANN is a subset of a general computing model known as connectionism. A connectionistic model
features an interconnected network of simple units that produce emergent properties that are beyond
the capabilities of the individual units (Medler, 1998). In the emergent respect, that the proverbial
whole is greater than the sum of its parts, an ANN is similar to a cellular automaton.
A prevalent type of ANN is the multilayer perceptron (MLP). MLPs are organized in layers, shown in
Figure 6. Layers are made up of a number of interconnected neurons. Patterns are presented to the
network via the input layer, which connects to one or more hidden layers where the actual processing is
done via a system of weights associated with the connections. The hidden layers then connect to an
output layer where the output is represented by the activation of one or more neurons.
Figure 6 – Multilayer perceptron. From (Cazala, 2015) with permission of Juan Cazala.
A weight value is associated with each connection in the network. The weights are multiplied by the
outputs of the source neurons, the sum of which is input to an activation function, which computes a
neuron’s output.
Activation functions are typically sigmoid shaped, such as the logistic function shown in Figure 7. Here
the output switches smoothly from 0 (off) to 1 (on) over an interval controlled by the β parameter. An
important property of an activation function is that is it differentiable, which allows a network to be
trained through learning.
Figure 7 – Logistic activation function. From (Sayed, 2016) with kind permission of Saed Sayed.
Learning is the process of modifying the connection weights to produce outputs that differ least from
“correct” outputs, i.e. minimized error. When the correct outputs are known, this type of learning is
called supervised learning. Learning typically entails backpropagating the output error from the output
to input neurons to modify the weights of the connections such that the error is reduced in subsequent
computations. The most common modification algorithm for this is the delta rule.
Training is a process that involves repetitive runs of input-output patterns through the network with an
application of the learning rule performed for each pattern. A run through an entire training set is called
an epoch. ANNs, like their biological counterparts, are known for their ability to produce correct outputs
given noisy or similar inputs. After training, a separate test set of patterns with input variations can be
evaluated to assess the effectiveness of training.
There are a number of ANN variations. For example, the activation function can be a Gaussian function
instead of a sigmoid one. ANNs with many layers are also possible. In general these are known as deep
learning networks. In a convolutional network, an input layer neuron feeds only a subset of neurons in
the next layer. This resembles the architecture of the human retina. An important ANN variation,
capable of learning input-output sequences such as those found in speech patterns, are called recurrent
ANNs.
Implementation
By squashing the morphogen sector values into an input vector and considering the cell types as a set of
outputs, an ANN can be trained to learn a set of metamorphs such that a morphogen derived from a cell
in a CA will map to the metamorph closest matching it. This is shown in Figure 8.
The advantages of using an ANN are threefold:
1. Speed: Instead of searching a set of metamorphs during execution for the closest morphogen,
an ANN performs a cascading set of arithmetic calculations to arrive at an output.
2. Compact representation: An ANN is capable of retaining a large number of input-output
mappings in the form of interconnection weights.
3. Generalization capability: An ANN is capable of classifying inputs that are similar to training
inputs. This capability will be exploited in subsequent applications.
Figure 8 – Neural network implementation.
Results
Various features of Morphozoic are illustrated by the following applications.
Conway’s Game of Life
This well-known CA (Gardner, 1970) was chosen as a baseline capability test for Morphozoic. In the
Game of Life (GoL), cells are in a rectangular array on a “game board”. Each cell is either in an “alive” or
“dead” state. The state change rules are as follows:
1. Any live cell with fewer than two live neighbors dies, as if caused by under-population.
2. Any live cell with two or three live neighbors lives on to the next generation.
3. Any live cell with more than three live neighbors dies, as if by overcrowding.
4. Any dead cell with exactly three live neighbors becomes a live cell, as if by reproduction.
A GoL “game” starts with an initial configuration of alive and dead cells. The rules are then applied to
the cells at each time step. Usually, if the initial configuration is random, the pattern of live and dead
cells appears to change chaotically for a while, then resolves itself into isolated clusters of cells that
cycle through a repeating series of states.
Figure 9 – A Game of Life configuration.
A sample GoL configuration is shown in Figure 9. Despite its simple rules, the GoL produces many
dynamic self-sustaining patterns (and is capable of producing an unlimited number of such patterns).
One class of pattern, called “gliders”, move across the game board as they go through their state
changes and interact in various ways with other clusters when collisions occur (Rendell, 2002). More
complex configurations called “guns” are a special type of self-sustaining pattern that produce gliders at
a specific rate. Using gadgets such as these it has been shown that a single-tape Turing machine can be
simulated by GoL (Rendell, 2002). In complexity theory, a set of rules that manipulates the state of a
system is described as Turing complete if it can simulate any single-tape Turing Machine. To
demonstrate the Turing completeness of another computational system all one needs to do is show that
it is capable of simulating an existing Turing complete system.
Examples of Turing complete systems include most commonly used programming languages, which span
paradigms such as procedural (C), functional (Haskell), and object-oriented (Java). A language’s Turing
completeness is determined by whether it has the ability to branch conditionally and load/store an
arbitrary number of variables. These two features, inherent to general purpose languages, can be
implemented in other paradigms using analogous structures such as using recursion in Haskell to
implement repetition. Turing completeness research has resulted in the discovery of more and more
simple systems that hold these properties. In fact, GoL is not the only cellular automata that has been
proven to be Turing complete nor is it the simplest.
The elementary cellular automaton Rule 110, whose state transition rules are stated in Table 1, has also
been shown to be Turing complete (Cook, 2004). Unlike GoL, Rule 110 functions in only one dimension,
each cell’s next state results only from the current states of its two neighbors.
B0 B1 B2
111
110
101
100
011
010
001
000
B1’
0
1
1
0
1
1
1
0
Table 1. State transition from B1 to B1’ given all possible starting states.
While proving that simple cellular automatons such as GoL and Rule 110 are Turing complete does not
allow us to compute in novel or more efficient ways, it does allow us to demonstrate the Turing
completeness of other systems such as Morphozoic. Proving that Morphozoic is Turing complete is
significant because “Turing completeness” means that it is a universal computer. Given enough memory,
Morphozoic would be able to perform any computation that any other Turing complete system is
capable of. This says nothing about how efficiently the computation will be done; it only says that the
computation can be done.
For Morphozoic, we “reverse engineer” a complete set of GoL rules, consisting of 512 3x3 Moore
neighborhood configurations. Generating a metamorph for each rule, the configurations were correctly
processed, both in the lookup and ANN implementations. Therefore, it follows that Morphozoic is also
Turing complete. From the perspective of modeling complex systems like embryos, this means that
Morphozoic is capable of modeling anything that can be computationally modeled.
Cell regeneration
The Morphozoic algorithm is capable of modeling cell regeneration. In order to demonstrate this, an
apparatus was devised that also highlights the functionality of nested neighborhoods. Figures 10 and 11
show the apparatus. The automaton is trained to regenerate either the horizontal bar on the right side
of Figure 10 or the vertical bar on the right side of Figure 11 beginning with the central block
configurations on the left side of the figures, respectively. The shaded borders are the distinguishing
features that determine regeneration direction: vertically shaded borders train for a horizontal bar and
horizontally shaded borders train for a vertical bar.
Figure 10 - Horizontal bar. Left: begin. Right: goal.
Figure 11 - Vertical bar. Left: begin. Right: goal.
To spotlight the results, only cells in the central 9x9 area in are allowed to create metamorphs during
training, and thus cells only in this area are allowed to regenerate by modifying their type values. Two
settings of the neighborhood dimension and number of neighborhoods were defined. The unnested
setting was a single 9x9 neighborhood. The nested setting was three nested 3x3 neighborhoods. The
unnested and nested automata thus equally contained 81 variable values. However, the nested
automaton neighborhoods span an area of 27 cells while the unnested automaton spans only 9 cells.
The nested automaton affords the extended range by aggregating cell values.
Figure 12 – nested morphogen fields: left=3x3, center=9x9, right=27x27
The automata were trained and tested for a variety of border lengths, which are manipulated by
changing the dimensions of the grid: the larger the grid, the longer its borders. And as the border
lengths increase, the distances from the central block of regenerating cells to the borders increase,
eventually passing out of range of the morphogenetic fields associated with the metamorphs. This can
be seen in Figure 12, which from left to right shows the morphogenetic field sizes for the 3x3, 9x9, and
27x27 nested neighborhoods. It can be seen that only the 27x27 neighborhood intersects with the left-
right border, allowing morphogens associated with the central cells to generate the correctly aligned
bar. Morphogens associated with the 3x3 and 9x9 neighborhoods do not intersect the border cells, and
thus cannot determine the correct orientation.
For each border length setting, the automata were sequentially exposed to the correctly oriented bars
for the vertical and horizontal borders, causing the creation of metamorphs that generate appropriate
cell types. A test is correspondingly in two parts: the central block with vertical borders followed by the
central block with horizontal borders. A successful test is defined as the regeneration of the correct bars
for both parts.
The results are shown in Table 2 for border length increments of four cells. Both the unnested and
nested automata correctly process the first two border lengths. However, at the third length, the
unnested automaton cannot detect the borders and thus fails to regenerate both bars correctly.
Eventually the border recedes beyond the range of the nested automaton as well, indicated in the last
row of the table.
Table 2 - Cell regeneration results.
The cell regeneration presented here can also be understood as a type of tropism, wherein a directional
movement or growth takes place in response to an environmental stimulus. Phototropism, for example,
is a response to light (Goyal, Szarzynska & Fankhauser, 2013). Another biological counterpart involving
cell (re)generation stimulated by fields of chemical signals is paracrine signaling for wound
healing (Hocking & Gibran, 2010; Dittmer & Leyh, 2014). Stem cells migrating to wound sites release
bioactive factors that orchestrate wound healing.
When tissues are wounded, they are healed via a regenerative process called epithelialization (Hocking
& Gibran, 2010). During this process, it has been found that stem cells, particularly mesenchymal stem
cells (MSCs) aid in the speed of wound healing by releasing growth factors and other extracellular
factors that facilitate regeneration (Hocking & Gibran, 2010). In general, stem cells are functionally
plastic, and can differentiate into a variety of functional cell lineages depending on where they are
recruited and the functional context. During organ regeneration, bone marrow stem cells migrate to
sites that are damaged to provide various roles in the regeneration process (Morigi et al., 2004). It is this
functional context where the parallels with Morphozoic's generative process can be found. For example,
MSCs often act through paracrine signaling rather than direct replacement of the cells making up
damaged tissue (Bruno et al., 2013) .
Stem cells also have the capacity to self-renew, which allow them to proliferate and thus persist for
many more divisions than differentiated cells. Perhaps more importantly, stem cells can manipulate
their local environments to favor regeneration (Dittmer & Leyh, 2014). This is done by acquiring the
gene expression patterns of their new defined fates, as well as producing a host of secretion factors. It is
this secretome (Salgado & Gimble, 2013) that most strongly influence the extracellular environment,
and can act to coordinate the behavior of cells and tissues at multiple scales. Influence of the local
environment is also accomplished via paracrine signaling (Baraniak & McDevitt, 2010), which involves
short-range chemical signaling of growth factors between cells. In a very simple manner, Morphozoic
can mimic these signaling mechanisms, and the architecture of Morphozoic could be used to implement
these mechanisms at multiple spatial scales.
Cell regeneration is concerned with the restoral of missing information using nearby available
information. A related process, digital image inpainting (DII) is a computer algorithm that restores
missing information of images such as those of old oil paintings. A biological counterpart occurs in
human visual systems as well in the form of blind spots (Satoh, 2012). As an illustration of the
Morphozoic cell regeneration capability applied to image restoral, consider the problem of restoring an
image from its edges, as shown in Figure 13.
Figure 13 – Lenna image with edges (Wikipedia, 2016a).
This is done by casting the original image and edge-detected image into a CA grid. The edge cells, shown
in the upper left part of Figure 14 (in reverse color for contrast), form an outline of the source pattern.
The original image, shown in the bottom right of Figure 14, forms the target pattern. Only the dark edge
cells are allowed to morph into target cells. With this restriction, five morphing steps are necessary to
complete the transition.
Figure 14 – Step-by-step image restoral from edges.
Gastrulation evolution
Background
There have been a number of attempts to model gastrulation and related developmental processes.
Developmental phenomena such as cleavage, blastulation, and gastrulation have been modeled using a
physical cellular model that indirectly assumes how genetic mechanisms effect the timing of these
processes (Drasdo and Forgacs, 2000). A more direct technique would involve the use of Genetic
Algorithms (GAs). GAs are a type of evolutionary computation that focuses on the construction and
optimization of programs using natural selection. In this case, natural selection act to select programs
that meet or exceed the criteria to reproduce and/or maintained in a population of programs. GAs have
been used in conjunction with Young's cellular automata to approximate the reaction-diffusion driven
biological pattern formation and replicate morphogenetic fields (Gravan and Lohoz-Beltra, 2004).
Because they are an instance of adaptive computing, there is the potential for broader application of
GAs to problems of biological development As an optimization procedure, GAs provide an adaptive
method that can find solutions despite a rugged search landscape (Bornholdt, 1998). This is analogous to
how biological populations evolve solutions to adaptive problems in a de novo fashion. We can
successfully approximate complex, emergent phenomena in development specifically and biology more
generally by more closely examining the relationship between computational representation and
biological complexity.
A GA consists of three parts: a genetic representation, a set of operators for mutation and/or crossover
operator, and a selection criterion (Holland, 1992). The algorithmic representation (genome) often
consists of one or more chromosomes, while each chromosome consists of several genes. A genetic
representation is a computational abstraction of a genome, each gene representing a compact encoding
of some aspect or feature in a given system. Each gene consists of serial bits and acts as a bit register,
which can be used to encode a wide variety of problems. This allows us to capture the power of
population dynamics and heredity rather than expression of genes, although specialized genetic
algorithms (Ferreira, 2001) can incorporate gene expression. In general, using genetic algorithms as a
form of theoretical inquiry allows us to replicate the dynamics of a given biological system, which in turn
allows us to better understand the complexity of the underlying system (Steventon and Arias, 2016).
The essential component of a genetic algorithm is the mutation and/or crossover operations. While
mutation consists of bit flipping, recombination consists of exchanging entire blocks of bits either within
or between programs. In both cases, an operator introduces variation in the function of a program. This
variation is then selected upon based on fitness criterion, but is also subject to historical constrains of
the problem space. This allows for the self-organization of patterns to emerge (Nizam and
Shanmugham, 2013) without breaking the code or destroying the structure of the initial program. One
way in which GAs avoid code fragility is to start with a population of programs that exhibit differential
reproduction. In this way, a genetic algorithm selects from a variety of potential solutions. A related
issue is the maintenance of structure and complexity in a naturally-selected program space is the
existence of building blocks (Forrest and Mitchell, 2014). This is similar to modularity in biological
evolution, in which parts of the program are protected from future evolutionary change (Wagner and
Altenberg, 1996).
All of these issues figure prominently into the approximation of gastrulation as a developmental
process. GAs are particularly good at finding heuristic solutions to problems residing in a large possibility
space, so they are likewise suited to simulating instances of biological self-organization where an exact
solution is not required. This is particularly true of simulating large problem spaces for where biological
experimentation would yield no clear solution. The problem of emergence in biological self-organization
can be viewed in terms of computational complexity (Grover, 2011). In gastrulation, we can see the
usefulness of both properties, in addition to the usefulness of incorporating gene expression into a
multi-scale model (Kaandorp et.al, 2012). Biological self-organization also relies on rules and constraints
rather than explicit instructions.
Implementation
Gastrulation is a process by which the cells of an embryo form an invagination in an early spheroid
shape in the course of tissue differentiation. A simple Morphozoic model of this is shown in Figure 15.
This configuration was formed through a progression of cell “divisions” starting from a single central cell.
Morphozoic generated metamorphs from a programmatically produced sequence. Two nested 3x3
neighborhoods were required; one neighborhood was found to be insufficient to reproduce the desired
output. Once generated, the metamorphs executed the sequence correctly.
Figure 15 – Gastrulation configuration.
A model that purports to explain the workings of morphogenesis at any level of abstraction should be
evolvable. In order to demonstrate the mutability and evolvability of Morphozoic, a genetic algorithm
(GA) was used to evolve a population of “organisms” to gastrulate successfully. The fitness function was
a count of the number of matching cell types summed over the sequence of cell type transitions.
Each population member contained metamorphs that were initially constructed by randomly pairing
morphogens generated from a programmed gastrulation sequence with actions from the sequence. The
number of metamorphs in each member was determined by the number required to execute the
sequence.
After a subset of the population was selected as fit members, the population was replenished with
mutants and offspring of fit members. Offspring were created by mating randomly chosen parents
members and performing a crossover operation on their metamorph sets. Crossover consisted of
supplying the child with a set of metamorphs randomly selected from its parents. Mutants were created
from individual fit members by discarding all metamorphs that were not executed and replacing them
with random ones.
Additional GA parameters:
● Population size = 50
● Fit population size = 10
● Number of generations = 50
● Number of offspring from matings = 20
Figure 16 – Gastrulation evolution results.
As Figure 16 indicates, the GA produced the required gastrulation morph sequence. The computation
was done in approximately 12 hours on standard desktop computer.
Neuron pathfinding
Also called axon guidance, neuron pathfinding is a process by which axons are guided by chemical
signals to target neurons, a process essential for the formation of organized neural networks (Tessier-
Lavigne & Goodman, 1996). While cellular automata have been used to simulate similar “branching”
phenomena (Markus, Böhm & Schmick, 1999), the generalization of reverse engineered rules is a novel
application.
One of the benchmarks presented here is that of neuron pathfinding. Morphozoic is able to accomplish
this gradually by exploiting the organizational feedback of cellular growth and spatial gradients
embodied in the metamorphs. In biology, neuron pathfinding is accomplished a bit differently, instead
using chemotropic mechanisms to guide axons to their target. Tropic cues are directional cues initiated
by various stimuli. In the case of chemotropic cues, chemical signals and gradients (e.g. the axonal
growth cone) guide regenerating tissues in the direction of its target (Huber et al., 2003). The
chemotropic mode of action has not only been shown to exist in developmental
morphogenesis (Tessier-Lavigne et al., 1988), but during tissue regeneration as well (Alto et al., 2009).
To simulate this, three types of cells are used: source neuron, target neuron, and axon. These can be
seen in Figure 17. In this application, source neurons are allowed to have multiple axons. Metamorphs
were generated from programmatic sequences of axon growth. For execution, however, the
metamorphs were used to train an ANN (see Figure 8), which classified actions based on pattern-
matching morphogens.
Figure 17 – Neuron pathfinding example.
The training set was created by growing axons from source to target neurons. The source and target
neurons were randomly placed along the left and right sides of the grid, respectively. Metamorphs were
then generated from multiple random configurations.
For testing, source and target neurons were randomly placed, morphogens generated from cell
neighborhoods, and axons grown by pattern-matching the learned patterns. The challenge here is that
axons must be grown from source to targets in different positions than those for which training was
done, so training must generalize to handle novel neuron positions. The employment of an ANN as a
classification mechanism was useful to accomplish this.
This procedure was done for a single source neuron, one and two target neurons, and with one, five,
and ten training set exemplars. A successful trial is defined as axons connecting to the target neurons.
Ten trials were run for each parameter configuration.
The results are shown in Figure 18. As might be expected, performance for the single target neuron
exceeded that for two targets, the latter having a greater number of possible configurations. The
number of exemplars also had a significant beneficial effect: one exemplar was insufficient to produce
any successes for the two target neuron case.
Figure 18 – Neuron pathfinding results.
Turing’s reaction-diffusion morphogenesis
Alan Turing is credited with pioneering a mathematical formulation of morphogenesis generally known
as a reaction-diffusion system (Turing, 1952). This system consists of a set of dynamically coupled
substances that, depending on parameters, are capable of producing various complex patterns, such as
stripes, spots, and spirals. Figure 19 shows a “cheetah coat” pattern generated from a random initial
pattern of three cell types.
Figure 19 – Reaction-diffusion morphogenesis of a “cheetah coat” pattern from random values.
This application is a comparison with this well-known specialized morphogenesis algorithm.
Metamorphs were generated from the Turing reaction-diffusion morphogenesis. These were used to
train an ANN to learn appropriate cell type changes. To test, a freshly randomized pattern was
generated and the metamorph pattern-matching mechanism executed. A typical result is shown in
Figure 20. While not as smooth appearing as the original, the spotted pattern is distinctly visible.
Figure 20 – Simulation of reaction-diffusion morphogenesis.
Conclusions
Biology presents us with numerous cases of morphogenetic fields as a morphogenesis mechanism. The
particular mechanisms and the depth of knowledge of these cases varies widely. Morphozoic is an
attempt to implement an abstraction of the functional commonality of morphogenetic fields as a
mechanism for self-organizing computation.
Morphozoic is a novel embodiment of a number of capabilities that are useful for morphogenesis:
flexibility, compact and economical computability, evolvability, and generalizability, especially in
association with the artificial neural network implementation. The cited applications were chosen to
demonstrate these properties.
Future directions
● Neighborhood sector value: The assignment of a value to a sector could be any function of the
types of its cellular components: average, mode, winner take all, etc. An alternative is to look at the
change of cell type as an image processing operation, such as taking a Laplacian, Sobel and other
edge enhancements, starbyte transformations (Sivaramakrishna & Gordon, 1997; Sivaramakrishna,
1998), contrast enhancement (Gordon & Rangayyan, 1984b, a), etc.
● Field signal strength and cell type variability: If cell types were continuous quantities instead of
discrete, a straightforward proportional mapping of morphogen similarity to cell type value would
be possible. On a related, more subtle possibility, morphogenetic fields are signaling constructs.
Signals can vary in some ways, such as amplitude, while retaining invariant signatures, such as
frequency spectra for electromagnetic waves. This suggests that variable action potentials
proportional to field strength could be a fruitful topic for research. For example, might scaled,
fractal-like structures be morphed?
● Dynamic/temporal fields: Limiting morphogenetic fields exclusively to spatial representations
restricts the model to a state machine. Dynamic fields that incorporate temporal information could
also be explored (Portegys & Wiles, 2004; Martínez, Adamatzky & Alonso-Sanz, 2013). This could
take the form of metamorphs that contain past neighborhood patterns or hierarchical streams of
neighborhoods that embody context.
● Fluid three-dimensional fields: A major reason that the cellular automaton architecture was chosen
was that it provides a straightforward mapping of local and global morphogenetic fields. However, a
fixed two dimensional grid is not a crucial feature of the Morphozoic model. The fluid three
dimensional medium that biological systems operate in points to an opportunity for the model to
explore.
Relevance to artificial intelligence
Morphozoic is also relevant for artificial intelligence (AI). AI research to a great degree focuses on the
brain and behaviors that the brain generates. But the brain, an extremely complex structure resulting
from millions of years of evolution, can be viewed as a solution to problems posed by the environment.
There is a common and somewhat ironic tendency to describe AI inputs and outputs in human cognitive
terms, i.e. post-processed brain output, such as symbolic variables (Hoffman, 2009).
An alternative approach, suggested by morphogenesis, is to view the environment as set of local/global,
spatial/temporal signal fields, and that the processing of fields is what spurs brain development.
Organisms are capable of performing amazing feats, such as navigation and nest-building, by the sensing
of unique environmental signals, such as polarized light, magnetism, and chemical scent trails.
Morphogenesis makes plain that signal fields can have powerful computing capabilities. Perhaps it is
worth exploring artificial intelligence as a solution to environments composed of these fields. As
Edmund Sinnott, author of Plant Morphogenesis (Sinnott, 1960), suggested, morphogenesis may be the
key to understanding intelligence (Sinnott, 1961; Sinnott, 1962b; Sinnott, 1962a; Sinnott, 1966), starting
with plant intelligence (Mancuso & Viola 2015).
In keeping with nature’s penchant for extending rather than replacing, the sponge-like shape of the
mammalian neocortex can be seen as symbolically apropos. For its purpose might be to soak up signals
from far reaches of time and space and render them, as though yet near and present, to the old brain
whose instinctual role has little changed over eons. The environmental gradients that clearly drive the
behavior of simpler creatures then became internalized in the nervous systems of more neurologically
complex ones.
The Morphozoic Java code is available here: https://github.com/pascualy/Morphozoic.
Morphozoic was developed in conjunction with these community projects that model and simulate the
C. elegans nematode worm:
● DevoWorm (Alicea et al., 2014; Alicea, 2016)
● OpenWorm (Szigeti et al., 2014)
References
Abbott, R. (2006). Emergence explained: Abstractions - Getting epiphenomena to do real work. Complexity 12(1), 13-26.
Adamatzky, A. (1997). Automatic programming of cellular automata: Identification approach. Kybernetes 26(2-3), 126-&.
Adami, C. (1998). Introduction to Artificial Life. New York, Springer.
Alberts, B., D. Bray, J. Lewis, M. Raff, K. Roberts & J.D. Watson (2002). Universal mechanisms of animal development. in:
Molecular Biology of the Cell. New York, Garland Publishing: http://www.ncbi.nlm.nih.gov/books/NBK26825/, 4th.
Alicea, B. & R. Gordon (2014). Toy models for macroevolutionary patterns and trends. BioSystems 122(Special Issue:
Patterns of Evolution), 25-37.
Alicea, B., S. McGrew, R. Gordon, S. Larson, T. Warrington & M. Watts. (2014). DevoWorm: differentiation waves and
computation in C. elegans embryogenesis. http://www.biorxiv.org/content/early/2014/10/03/009993
Alicea, B. (2016). DevoWorm Project. http://devoworm.weebly.com
Alto, L.T., L.A. Havton, J.M. Conner, E.R. Hollis, II, A. Blesch & M.H. Tuszynski (2009). Chemotropic guidance facilitates
axonal regeneration and synapse formation after spinal cord injury. Nat. Neurosci. 12(9), 1106-U1108.
Ashby, W.R. (1968). Some consequences of Bremermann's limit for information-processing systems. In: Problems in
Bionics. Ed.: H.L. Oestreicher & D.L. Moore. New York, Gordon and Breach Science Publishers, Inc.: 69-76.
Bandini, S. & G. Mauri (1999). Multilayered cellular automata. Theor. Comput. Sci. 217(1), 99-113.
Baraniak, P.R. & T.C. McDevitt (2010). Stem cell paracrine actions and tissue regeneration. Regen. Med. 5(1), 121-143.
Bateson, P., D. Barker, T. Clutton-Brock, D. Deb, B. D'Udine, R.A. Foley, P. Gluckman, K. Godfrey, T. Kirkwood, M.M. Lahr,
J. McNamara, N.B. Metcalfe, P. Monaghan, H.G. Spencer & S.E. Sultan (2004). Developmental plasticity and human
health. Nature 430(6998), 419-421.
Batten, D.F. (2001). Complex landscapes of spatial interaction. Annals of Regional Science 35(1), 81-111.
Batty, M. (2003). Geocomputation using cellular automata. In: Geocomputation. Ed.: R.J. Abrahart, S. Openshaw & L.M.
See, CRC Press: 95-126.
Beloussov, L.V., J.M. Opitz & S.F. Gilbert (1997). Life of Alexander G. Gurwitsch and his relevant contribution to the
theory of morphogenetic fields. Int J Dev Biol 41(6), 771-779.
Beloussov, L.V. (2015). Morphogenetic fields: History and relations to other concepts. In: Fields of the Cell. Ed.: D. Fels,
M. Cifra & F. Scholkmann. Kerala, India, Research Signpost: 271-282.
Bijelic, A., C. Molitor, S.G. Mauracher, R. Al-Oweini, U. Kortz & A. Rompel (2015). Hen egg-white lysozyme crystallisation:
protein stacking and structure stability enhanced by a tellurium(VI)-centred polyoxotungstate. ChemBioChem 16(2), 233-
241.
Bornholdt, S. (1998). Genetic algorithm dynamics on a rugged landscape. Physical Review E, 57(4), 3853.
Bourgine, P. & A. Lesne, Ed. (2010). Morphogenesis: Origins of Patterns and Shapes. Berlin, Springer Science & Business
Media.
Bower, J.M. (2005). Looking for Newton: realistic modeling in modern biology. Brains, Minds and Media 1(2), bmm217.
Bremermann, H.J. (1967). Quantum noise and information. In: Proceedings of the Berkeley Symposium on Mathematical
Statistics and Probability. Ed. Berkeley, University of California Press. 4: 15-22.
Bruno, S., F. Collino, C. Tetta & G. Camussi (2013). Dissecting paracrine effectors for mesenchymal stem cells. In:
Mesenchymal Stem Cells: Basics and Clinical Application I. Ed.: B. Weyand, M. Dominici, R. Hass, R. Jacobs & C. Kasper.
129: 137-152.
Bubak, M. & P. Czerwiński (1999). Traffic simulation using cellular automata and continuous models. Computer Physics
Communications 121, 395-398.
Burguillo, J.C. (2013). Playing with complexity: From cellular evolutionary algorithms with coalitions to self-organizing
maps. Computers & Mathematics with Applications 66(2), 201-212.
Cazala, J. (2015). Perceptron. https://github.com/cazala/synaptic/wiki/Architect
Chopard, B. (1990). A cellular automata model of large-scale moving objects. Journal of Physics A-Mathematical and
General 23(10), 1671-1687.
Cook, M. (2004). Universality in elementary cellular automata. Complex systems 15(1), 1-40.
Dascalu, M., G. Stefan, A. Zafiu & A. Plavitu (2011). Applications of multilevel cellular automata in epidemiology. In:
Proceedings of the 13th WSEAS international conference on Automatic control, modelling & simulation. Ed., World
Scientific and Engineering Academy and Society (WSEAS): 439-444.
de Koster, C.G. & A. Lindenmayer (1987). Discrete and continuous models for heterocyst differentiation in growing
filaments of blue-green bacteria. Acta Biotheoretica 36(4), 249-273.
Deutsch, A., P. Maini & S. Dormann, Ed. (2007). Cellular Automaton Modeling of Biological Pattern Formation:
Characterization, Applications, and Analysis, Birkhäuser Boston.
Dittmer, J. & B. Leyh (2014). Paracrine effects of stem cells in wound healing and cancer progression (Review). Int. J.
Oncol. 44(6), 1789-1798.
Dobrescu, R. & V.I. Purcarea (2011). Emergence, self-organization and morphogenesis in biological structures. Journal of
medicine and life 4(1), 82-90.
Drachman, D.A. (2005). Do we have brain to spare? Neurology 64(12), 2004-2005.
Drasdo, D. and Forgacs, G. (2000). Modeling the Interplay of Generic and Genetic Mechanisms in Cleavage, Blastulation,
and Gastrulation. Developmental Dynamics, 219(2), 182-191.
Dunn, A. (2010). Hierarchical cellular automata methods. In: Simulating Complex Systems by Cellular Automata. Ed.: A.G.
Hoekstra, J. Kroc & P.M.A. Sloot: 59-80.
Dunn, A.G. & J.D. Majer (2007). Simulating weed propagation via hierarchical, patch-based cellular automata. Lecture
Notes in Computer Science 4487, 762-769.
Elmenreich, W. & I. Fehervari (2011). Evolving self-organizing cellular automata based on neural network genotypes.
Lecture Notes in Computer Science 6557, 16-25.
Ferrando, N., M.A. Gosálvez, J. Cerdá, R. Gadea & K. Sato (2011). Octree-based, GPU implementation of a continuous
cellular automaton for the simulation of complex, evolving surfaces. Computer Physics Communications 182(3), 628-640.
Ferrando, N., M.A. Gosálvez & R.J. Colóm (2012). Evolutionary continuous cellular automaton for the simulation of wet
etching of quartz. Journal of Micromechanics and Microengineering 22(2).
Ferreira, C. (2001). Gene expression programming: a new adaptive algorithm for solving problems. Complex Systems,
13(2), 87-129.
Forrest, S. and Mitchell, M. (2014). Relative Building-Block Fitness and the Building Block Hypothesis. In Foundations of
Genetic Algorithms II. Morgan Kaufmann, San Mateo, CA. pp. 109-126.
Fredkin, E. (1992). Finite nature. Progress in Atomic Physics Neutrinos and Gravitation 72, 345-354.
Fukui, M. & Y. Ishibashi (1996). Traffic flow in 1D cellular automaton model including cars moving with high speed.
Journal of the Physical Society of Japan 65(6), 1868-1870.
Gardner, M. (1970). Mathematical Games: The fantastic combinations of John Conway's new solitaire game "life".
Scientific American 223(4, October), 120-123.
Ghobara, M., D.R. Smith, B. Schoefs, V. Vinayak, I.C. Gebeshuber & R. Gordon (2016). On light and diatoms: A photonics
and photobiology review. Advances in Optics and Photonics, to be submitted.
Gluckman, P.D., M.A. Hanson & F.M. Low (2011). The role of developmental plasticity and epigenetics in human health.
Birth Defects Research Part C-Embryo Today-Reviews 93(1), 12-18.
Gordon, N.K. & R. Gordon (2016a). Embryogenesis Explained. Singapore, World Scientific Publishing.
Gordon, N.K. & R. Gordon (2016b). The organelle of differentiation in embryos: the cell state splitter [invited review].
Theoretical Biology and Medical Modelling 13(Special issue: Biophysical Models of Cell Behavior, Guest Editor: Jack A.
Tuszynski), #11.
Gordon, R. (1966). On stochastic growth and form. Proceedings of the National Academy of Sciences USA 56(5), 1497-
1504.
Gordon, R. (1970). On Monte Carlo algebra. Journal of Applied Probability 7, 373-387.
Gordon, R., N.S. Goel, M.S. Steinberg & L.L. Wiseman (1972). A rheological mechanism sufficient to explain the kinetics of
cell sorting. J. Theor. Biol. 37(1), 43-73.
Gordon, R., N.S. Goel, M.S. Steinberg & L.L. Wiseman (1975). A rheological mechanism sufficient to explain the kinetics of
cell sorting. In: Mathematical Models for Cell Rearrangement. Ed.: G.D. Mostow. New Haven, Yale University Press: 196-
230.
Gordon, R. (1980). Monte Carlo methods for cooperative Ising models. In: Cooperative Phenomena in Biology. Ed.: G.
Karreman. New York, Pergamon Press: 189-241 + errata.
Gordon, R. & R.M. Rangayyan (1984a). Feature enhancement of film mammograms using fixed and adaptive
neighborhoods. Applied Optics 23(4), 560-564.
Gordon, R. & R.M. Rangayyan (1984b). Correction: Feature enhancement of film mammograms using fixed and adaptive
neighborhoods. Applied Optics 23(13), 2055.
Gordon, R. (1999). The Hierarchical Genome and Differentiation Waves: Novel Unification of Development, Genetics and
Evolution. Singapore & London, World Scientific & Imperial College Press:
[http://www.worldscientific.com/worldscibooks/10.1142/2755 ].
Gordon, R. & C.A. Melvin (2003). Reverse engineering the embryo: a graduate course in developmental biology for
engineering students at the University of Manitoba, Canada. International Journal of Developmental Biology 47(2/3),
183-187.
Gordon, R. (2015). Walking the tightrope: the dilemmas of hierarchical instabilities in Turing’s morphogenesis [invited].
In: The Once and Future Turing: Computing the World. Ed.: S.B. Cooper & A. Hodges. Cambridge, Cambridge University
Press: 150-164.
Gordon, R. & R. Stone (2016). Cybernetic embryo. In: Biocommunication. Ed.: R. Gordon & J. Seckbach. London, World
Scientific Publishing: in press.
Gosálvez, M.A., Y. Xing, K. Sato & R.M. Nieminen (2009). Discrete and continuous cellular automata for the simulation of
propagating surfaces. Sensors and Actuators A-Physical 155(1), 98-112.
Goyal, A., B. Szarzynska & C. Fankhauser (2013). Phototropism: at the crossroads of light-signaling pathways. Trends Plant
Sci. 18(7), 393-401.
Gravan, C.P. and Lohoz-Beltra, R. (2004). Evolving morphogenetic fieds in the zebra skin pattern based on Turing's
morphogen hypothesis. International Journal of Applied Mathematics and Computer Science, 14(3), 351-361.
Grover, M. (2011). Parallels between Gluconeogenesis and Synchronous Machines. International Journal on Computer
Science and Engineering, 3(1), 185-191.
Guan, Q. & K.C. Clarke (2010). A general-purpose parallel raster processing programming library test application using a
geographic cellular automata model. International Journal of Geographical Information Science 24(5), 695-722.
Halbach, M. & R. Hoffmann (2005). Optimal behavior of a moving creature in the cellular automata model. Lecture Notes
in Computer Science 3606, 129-140.
Haykin, S.O. (2011). Neural Networks and Learning Machines, Pearson Education, 3rd.
Hochberger, C., R. Hoffmann & S. Waldschmidt (1999). CDL++ for the description of moving objects in cellular automata.
Lecture Notes in Computer Science 1662, 428-435.
Hocking, A.M. & N.S. Gibran (2010). Mesenchymal stem cells: Paracrine signaling and differentiation during cutaneous
wound repair. Experimental Cell Research 316(14), 2213-2219.
Hodges, A. (2014). Alan Turing: The Enigma. New York, Simon & Schuster, 2nd.
Hoffman, D.D. (2009). The interface theory of perception: Natural selection drives true perception to swift extinction. In:
Object Categorization: Computer and Human Vision Perspectives. Ed.: S.J. Dickinson, A. Leonardis, B. Schiele & M.J. Tarr.
Cambridge, Cambridge University Press: 148-165.
Holland, J.H. (1992). Adaptation in Natural and Artificial Systems: An Introductory Analysis with Applications to Biology,
Control, and Artificial Intelligence. Cambridge, MA USA, MIT Press, revised.
Huber, A.B., A.L. Kolodkin, D.D. Ginty & J.F. Cloutier (2003). Signaling at the growth cone: Ligand-receptor complexes and
the control of axon growth and guidance. Annual Review of Neuroscience 26, 509-563.
Huberman, B.A. & N.S. Glance (1993). Evolutionary games and computer simulations. Proceedings of the National
Academy of Sciences of the United States of America 90(16), 7716-7718.
Ilachinski, A. (2001). Cellular Automata: A Discrete Universe.
http://www.worldscientific.com/worldscibooks/10.1142/4702
Jacob, F. (1977). Evolution and tinkering. Science 196(4295), 1161-1166.
Johnson, S. (2010). Where Good Ideas Come From, Penguin Publishing Group.
Kaandorp, J.A., Botman, D., Tamulonis, C., and Dries, R. (2012). Multi-scale Modeling of Gene Regulation of
Morphogenesis. In How the World Computes: Turing Centenary conference and 8th conference on computability in
Europe. S.B. Cooper, A. Dawar, and B. Lowe, eds. Lecture Notes in Computer Science, #7318. Springer, Berlin. pp. 355-
362.
Kaji, M. & T. Inohara (2014). Numerical analysis focused on each agent's moving on refuge under congestion
circumstances by using cellular automata. In: 2014 IEEE International Conference on Systems, Man and Cybernetics. Ed.:
2922-2927.
Katkovnik, V., K. Egiazarian & J. Astola (2006). Local Approximation Techniques in Signal and Image Processing.
Bellingham, WAashington, USA, SPIE Press.
Kier, L.B., D. Bonchev & G.A. Buck (2005). Modeling biochemical networks: A cellular-automata approach. Chem.
Biodivers. 2(2), 233-243.
Kiester, A.R. & K. Sahr (2008). Planar and spherical hierarchical, multi-resolution cellular automata. Computers
Environment and Urban Systems 32(3), 204-213.
Kitakawa, A. (2004). On a segregation intensity parameter in continuous-velocity lattice gas cellular automata for
immiscible binary fluid. Chemical Engineering Science 59(14), 3007-3012.
Kitakawa, A. (2005). Simulation of break-up behavior of immiscible droplet under shear field by means of continuous-
velocity lattice gas cellular automata. Chemical Engineering Science 60(20), 5612-5619.
Koestler, A. (1967). The Ghost in the Machine. New York, Macmillan.
Kohl, P. & D. Noble (2009). Systems biology and the virtual physiological human. Molecular Systems Biology 5.
Leavitt, D. (2006). The Man Who Knew Too Much: Alan Turing and the Invention of the Computer. New York, W. W.
Norton.
Levin, M. (2011). The wisdom of the body: future techniques and approaches to morphogenetic fields in regenerative
medicine, developmental biology and cancer. Regen. Med. 6(6), 667-673.
Levin, M. (2012). Morphogenetic fields in embryogenesis, regeneration, and cancer: non-local control of complex
patterning. Biosystems 109(3, SI), 243-261.
Li, Y., M.A. Gosálvez, P. Pal, K. Sato & Y. Xing (2015). Particle swarm optimization-based continuous cellular automaton
for the simulation of deep reactive ion etching. Journal of Micromechanics and Microengineering 25(5).
Liu, Y., H.D. Cheng, J.H. Huang, Y.T. Zhang & X.L. Tang (2012). An effective approach of lesion segmentation within the
breast ultrasound image based on the cellular automata principle. Journal of Digital Imaging 25(5), 580-590.
Lobo, D. & M. Levin (2015). Inferring regulatory networks from experimental morphological phenotypes: a computational
method reverse-engineers planarian regeneration. PLoS Comput Biol 11(6), e1004295.
Low, F.M., P.D. Gluckman & M.A. Hanson (2012). Developmental plasticity, epigenetics and human health. Evolutionary
Biology 39(4), 650-665.
Mancuso, S. & A. Viola (2015). Brilliant Green: The Surprising History and Science of Plant Intelligence, Island Press.
Markus, M., D. Böhm & M. Schmick (1999). Simulation of vessel morphogenesis using cellular automata. Math Biosci
156(1-2), 191-206.
Martínez, G.J., A. Adamatzky & R. Alonso-Sanz (2013). Designing complex dynamics in cellular automata with memory.
International Journal of Bifurcation and Chaos 23(10), 1330035.
Medler, D.A. (1998). A brief history of connectionism. Neural Computing Surveys 1, 18-72.
Moczek, A.P., S. Sultan, S. Foster, C. Ledon-Rettig, I. Dworkin, H.F. Nijhout, E. Abouheif & D.W. Pfennig (2011). The role of
developmental plasticity in evolutionary innovation. Proc. R. Soc. B-Biol. Sci. 278(1719), 2705-2713.
Mofrad, M.H., S. Sadeghi, A. Rezvanian & M.R. Meybodi (2015). Cellular edge detection: Combining cellular automata
and cellular learning automata. AEU-Int. J. Electron. Commun. 69(9), 1282-1290.
Monetti, R.A. & E.V. Albano (1997). On the emergence of large-scale complex behavior in the dynamics of a society of
living individuals: the Stochastic Game of Life. Journal of Theoretical Biology 187(2), 183-194.
Morigi, M., B. Imberti, C. Zoja, D. Corna, S. Tomasoni, M. Abbate, D. Rottoli, S. Angioletti, A. Benigni, N. Perico, M. Alison
& G. Remuzzi (2004). Mesenchymal stem cells are renotropic, helping to repair the kidney and improve function in acute
renal failure. J. Am. Soc. Nephrol. 15(7), 1794-1804.
Morozova, N. & M. Shubin. (2012). The Geometry of Morphogenesis and the Morphogenetic Field Concept.
http://arxiv.org/abs/1205.1158v1
Moussa, N. (2005). A 2-dimensional cellular automaton for agents moving from origins to destinations. Int. J. Mod. Phys.
C 16(12), 1849-1860.
Nizam, A. and Shanmugham, B. (2013). Self-organizing Genetic Algorithm: a survey. International Journal of Computer
Applications, 65(18), 0975-8887.
Noble, D. (2002). The rise of computational biology. Nat Rev Mol Cell Biol 3(6), 459-463.
Olami, Z., H.J.S. Feder & K. Christensen (1992). Self-organized criticality in a continuous, nonconservative cellular
automaton modeling earthquakes. Physical Review Letters 68(8), 1244-1247.
Opitz, J.M. & G. Neri (2013). Historical perspective on developmental concepts and terminology. American Journal of
Medical Genetics Part A 161(11), 2711-2725.
Portegys, T. & J. Wiles (2004). A robust game of life. In: The International Conference on Complex Systems (ICCS2004),
Boston, MA, USA, May 16-21, 2004. Ed.: http://www.necsi.edu/events/iccs/openconf/author/abstractbook.php.
Portegys, T.E. (2002). An abstraction of intercellular communication. In: Alife VIII Proceedings. Ed.
Rendell, P. (2002). Turing universality of the game of life. In: Collision-Based Computing. Ed.: A. Adamatzky. London,
Springer London: 513-539.
Resnick, M. (1994). Turtles, Termites, and Traffic Jams: Explorations in Massively Parallel Microworlds. Cambridge, MIT
Press.
Ruxton, G.D. (1996). Effects of the spatial and temporal ordering of events on the behaviour of a simple cellular
automaton. Ecological Modelling 84(1-3), 311-314.
Ruxton, G.D. & L.A. Saravia (1998). The need for biological realism in the updating of cellular automata models. Ecological
Modelling 107(2-3), 105-112.
Salgado, A.J. & J.M. Gimble (2013). Secretome of mesenchymal stem/stromal cells in regenerative medicine Foreword.
Biochimie 95(12), 2195.
Satoh, S. (2012). Computational identity between digital image inpainting and filling-in process at the blind spot. Neural
Computing & Applications 21(4), 613-621.
Sayad, S. (2016). Artificial Neural Network. http://www.saedsayad.com/artificial_neural_network.htm
Sinnott, E.W. (1960). Plant Morphogenesis. New York, McGraw-Hill Book Co.
Sinnott, E.W. (1961). Cell and psych. The biology of purpose.
Sinnott, E.W. (1962a). The Biology of The Spirit. New York The Viking Press.
Sinnott, E.W. (1962b). Matter, Mind and Man: The Biology of Human Nature. New York, Atheneum.
Sinnott, E.W. (1966). The Bridge of Life, From Matter to Spirit. New York, Simon and Schuster.
Sivaramakrishna, R. & R. Gordon (1997). Mammographic image registration using the Starbyte transformation. In:
WESCSANEX'97. Ed.: P.G. McLaren & W. Kinsner. Winnipeg, Canada, IEEE: 144-149.
Sivaramakrishna, R. (1998). Breast image registration using a textural transformation. Medical Physics 25(11), 2249.
Solé, R.V., S. Valverde, M. Rosas Casals, S.A. Kauffman, D. Farmer & N. Eldredge (2013). The evolutionary ecology of
technological innovations. Complexity 18(4), 15-27.
Steventon, B. and Arias, A.M. (2016). Evo-engineering and the cellular and molecular origins of the vertebrate spinal
cord. bioRxiv, doi:10.1101/068882
Sun, X., P.L. Rosin & R.R. Martin (2011). Fast rule identification and neighborhood selection for cellular automata. IEEE
Trans. Syst. Man Cybern. Part B-Cybern. 41(3), 749-760.
Szigeti, B., P. Gleeson, M. Vella, S. Khayrulin, A. Palyanov, J. Hokanson, M. Currie, M. Cantarelli, G. Idili & S. Larson (2014).
OpenWorm: an open-science approach to modelling Caenorhabditis elegans. Frontiers in Computational Neuroscience 8,
#137.
Tanaka, S. (2015). Simulation frameworks for morphogenetic problems. Computation 3(2), 197-221.
Tessier-Lavigne, M., M. Placzek, A.G.S. Lumsden, J. Dodd & T.M. Jessell (1988). Chemotropic guidance of developing
axons in the mammalian central nervous system. Nature 336(6201), 775-778.
Tessier-Lavigne, M. & C.S. Goodman (1996). The molecular biology of axon guidance. Science 274(5290), 1123-1133.
Tompkins, N., N. Li, C. Girabawe, M. Heymann, G.B. Ermentrout, I.R. Epstein & S. Fraden (2014). Testing Turing's theory
of morphogenesis in chemical cells. Proceedings of the National Academy of Sciences of the United States of America
111(12), 4397-4402.
Tria, F., V. Loreto, V.D.P. Servedio & S.H. Strogatz (2014). The dynamics of correlated novelties. Scientific Reports 4.
Turing, A.M. (1952). The chemical basis of morphogenesis. Phil. Trans. Roy. Soc. London B237, 37-72.
Tyldum, M. (2014). The Imitation Game [movie]. http://theimitationgamemovie.com
Tyler, S.E.B. (2014). The work surfaces of morphogenesis: The role of the morphogenetic field. Biological Theory 9(2),
194-208.
Ulam, S. (1962). On some mathematical problems connected with patterns of growth of figures. Symposium in Applied
Mathematics 14, 215-224.
van Nimwegen, E., J.P. Crutchfield & M. Huynen (1999). Neutral evolution of mutational robustness. Proceedings of the
National Academy of Sciences of the United States of America 96(17), 9716-9720.
Vecchi, D. & I. Hernández (2014). The epistemological resilience of the concept of morphogenetic field. In: Towards a
Theory of Development. Ed.: A. Minelli & T. Pradeu, Oxford University Press.
von Neumann, J. & A.W. Burks (1966). Theory of Self-Reproducing Automata. Urbana, University of Illinois Press.
Waddington, C.H. & R.J. Cowe (1969). Computer simulation of a mulluscan pigmentation pattern. J. Theor. Biol. 25, 219-
225.
Wagner, A. & W. Rosen (2014). Spaces of the possible: universal Darwinism and the wall between technological and
biological innovation. Journal of the Royal Society Interface 11(97).
Wagner, G.P. and Altenberg, L. (1996). Complex Adaptations and the Evolution of Evolvability. Evolution, 50(3), 967-976.
Walker, D.C. & J. Southgate (2009). The virtual cell-a candidate co-ordinator for 'middle-out' modelling of biological
systems. Brief. Bioinform. 10(4), 450-461.
Weimar, J.R. (2001). Coupling microscopic and macroscopic cellular automata. Parallel Computing 27(5), 601-611.
Weisstein, E.W. (2016a). Cellular Automaton. http://mathworld.wolfram.com/CellularAutomaton.html
Weisstein, E.W. (2016b). Moore Neighborhood. http://mathworld.wolfram.com/MooreNeighborhood.html
West-Eberhard, M.J. (2002). Developmental Plasticity and Evolution. Oxford, Oxford University Press.
Wikipedia. (2016a). Lenna. https://en.wikipedia.org/wiki/Lenna
Wikipedia. (2016b). List of animals by number of neurons.
https://en.wikipedia.org/wiki/List_of_animals_by_number_of_neurons
Wolfram, S. (1984). Cellular automata as models of complexity. Nature 311(5985), 419-424.
Wolfram, S. (2001). A New Kind of Science. Champaign, Illinois, Wolfram Media.
Wyczalkowski, M.A., Z. Chen, B.A. Filas, V.D. Varner & L.A. Taber (2012). Computational models for mechanics of
morphogenesis. Birth Defects Research Part C-Embryo Today-Reviews 96(2), 132-152.
Zhao, Y., H.L. Wei & S.A. Billings (2012). A new adaptive fast cellular automaton neighborhood detection and rule
identification algorithm. IEEE Trans. Syst. Man Cybern. Part B-Cybern. 42(4), 1283-1287.
