Download Morphogenesis and Replication of Multi

1
Morphogenesis and Replication
of Multi-Cellular Organisms with
Evolved Variable Length Self-Modifying Genomes
Stefano Nichele and Gunnar Tufte
ECAL 2015 – York, UK
July 20-24, 2015
Stefano Nichele, 2015
2
Inspiration / Motivation
Genomes of biological organisms are not fixed in size
They evolved and diverged into different species acquiring
new genes and thus having different lengths
•  LUA (Last Universal Ancestor) ~ 3.5 / 3.8 billion years ago
•  Gene duplication: redundant gene with less selection pressure
•  Larger genome: genetic novelty, potential for innovation
•  Complexification: incremental elaboration
•  ~ 38% Homo Sapiens genome due to gene duplication
3
Artificial morphogenetic systems often have static size genomes
System designer choice:
•  Trial & error
•  Estimation / heuristics
Fixed maximum complexity (vs. open-ended in nature)
4
Outline
•  CA as morphogenetic systems
•  Previous work:
–  Genome growth (attractors)
–  Morphogenesis and replication
•  Current / future work:
–  True complexification
–  Self-modifying genome (regulation)
–  Artificial stem cell mechanism
5
CA as morphogenetic systems
• 
CA can be considered as a morphogenetic system, in which an organism can
develop (e.g. grow) from a zygote to a multi-cellular organism (phenotype)
according to specific local rules, represented by a genome (genotype).
• 
The behavior of the CA is represented by the emergent phenotype, which is
subject to shape and size modification, along the developmental process.
6
Traditional CA model
Example CA with 4 cell states and 5 neighbors:
Search space = 4^4^5 = 41024 = ~ 3.23 x 10616
7
Previous work
S. Nichele, A. Giskeødegård and G. Tufte.
Evolutionary Growth of Genome Representations
on Artificial Cellular Organisms with Indirect
Encodings. Journal of Artificial Life. MIT Press.
ACCEPTED (2015)
•  Evolutionary growth of genomes
–  CA transition tables
–  abstract measure of phenotypic complexity: attractor length
•  Scalability
–  Search space, number of cell states, geomerty size (phenotypic resources)
• 
• 
• 
Allows speciation
Through gene duplication (as in nature)
Complexification (incremental elaboration)
Ø 
• 
NEAT (Stanley & Miikkulainen): good for evolving modular structures with direct encodings
Compare full vs restricted vs growing (genomes)
8
Regulation mechanisms:
•  Upper bound, duplication rate, optimization time, elitism
9
Scalability in search space – genome comparison
10
Scalability in state space
11
Scalability in solution space - geometry
12
Evolutionary growth of genome representations
• 
• 
• 
• 
• 
• 
Compact and effective genomes
Scalability of search space
Scalability of state space
Scalability of phenotypic resources
Start in low dimensional space
Incrementally increase genotype complexity
13
Previous work
S. Nichele and G. Tufte. Evolutionary Growth of
Genome for the Development and Replication of
Multicellular Organisms with Indirect Encodings.
IEEE SSCI, International Conference on Evolvable
Systems. (ICES 2014)
•  Morphogenesis and replication of different structures
•  Different mapping: IBD (Instruction-Based Development)
–  Not bounded (evolve from one instruction to program)
–  Traditional CA transition tables vs. growing genome with IBD
14
CA – IBD
(Bidlo and Skarvada 2008, Bidlo and Vasicek 2012)
U
L
C R
D
gene
Inst.
Code
Op1
Op2
15
Benchmark structures
16
Morphogenesis problem
6 1 3
3 3 1
8 3 4 11 4 3 6 3 0
2 4 3
1 1 4
2 1 4
6 4 1 15 4 3 5 2 0
1 2 4 13 4 0 0 2 0
•  Example of evolved program for the development of structure 2c – patch structure
•  After development step 9 the structure remains stable (point attractor)
•  The program is composed by 14 instructions (one instruction each gene)
•  INSTRUCTION CODE, OPERAND 1, OPERAND 2 (if the operand is not applicable for the given
instruction, the value is ignored)
•  Operands: UP = 0, RIGHT = 1, DOWN = 2, LEFT = 3, CENTRE = 4.
17
Morphogenesis- results
Replication problem
time
18
19
Replication - results
20
Success rate
Morphogenesis (avg. 100 runs)
Replication (avg. 100 runs)
Table-based Evolution
Success
Rate %
Table-based Evolution
Genotype Size (# genes)
Max
Avg
Min
StDev
Generations
Avg. StDev.
Success
Rate %
Genotype Size (# genes)
Max
Avg
Min
StDev
Generations
Avg. StDev.
A
58
32
32
32
0
1336
2294
A
85
32
32
32
0
775
1393
B
69
32
32
32
0
2254
2501
C
8
1024
1024
1024
0
4331
3576
C
19
1024
1024
1024
0
5002
3157
D
1
32
32
32
0
8259
0
D
23
32
32
32
0
2668
2942
E
0
1024
1024
1024
0
-
-
Instruction-based Growing Evolution
Success
Rate %
Instruction-based Growing Evolution
Genotype Size (# genes)
Max
Avg
Min
StDev
Generations
Avg. StDev.
Success
Rate %
Max
Genotype Size (# genes)
Avg
Min
StDev
Generations
Avg. StDev.
A
98
31
14.34
5
8.4318
1257
1152
A
100
7
2.93
2
1.1742
39.7
19.6
B
98
31
15.28
5
7.0973
3956
1690
C
100
6
2.84
2
1.1166
39.6
22.3
C
46
46
19.65
6
9.2236
6424
1922
D
100
8
3.06
2
1.2128
41.8
20.5
D
100
13
5.25
4
1.4097
285
108
E
100
5
1.38
1
0.8012
9.4
10.7
21
Evolutionary growth of genome - IBD
•  Initialize with single gene, allow duplication and
speciation
•  Traditional CA mapping vs instruction based
development (unbounded)
•  Morphogenesis and replication problems
•  Compact and effective genotype solutions (not
designed a priori)
•  Better success rate
22
Current / Future work
•  True complexification (in addition to genome growth)
–  Allow growth of available cell states
–  Unbounded state space (more or less fixed in every artificial system)
•  Introduce self-modifying instructions
– 
– 
– 
– 
Can modify genotype itself (genotype activation/regulation mechanism)
Allow diversification of cell programs
Allow hierarchical organization of cells (tissues, organs, organism)
Emergence of some kind of artificial stem cell mechanism
23
24
Initial results: comparison on the flag morphogenesis problem
25
Stem cell mechanism:
•  vital part of the complex
development process of any
multi-cellular organism
•  also serves in maintaining the
organism once fully developed
•  by definition replicators
1.  Genome complexification
2.  States complexification
3.  Genome self-modification
26
Stefano Nichele
www.nichele.eu