1 Emergent Evolutionary Dynamics Emergent Evolutionary Dynamics of Self-Reproducing Cellular of Self-Reproducing Cellular Automata Automata Chris Salzberg Chris Salzberg
Jan 19, 2016
1
Emergent Evolutionary Dynamics Emergent Evolutionary Dynamics of Self-Reproducing Cellular of Self-Reproducing Cellular
AutomataAutomata
Chris SalzbergChris Salzberg
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 2
CreditsCredits
Research for this project fulfills requirements for theResearch for this project fulfills requirements for the
Master of Science Degree - Computational ScienceMaster of Science Degree - Computational Science
Universiteit van AmsterdamUniversiteit van Amsterdam
Project work conducted jointly with Project work conducted jointly with Antony AntonyAntony Antony (SCS)(SCS)
Supervised by Supervised by Dr. Hiroki SayamaDr. Hiroki Sayama
(University of Electro-Communications, Japan)(University of Electro-Communications, Japan)
Mentor: Prof. Dick van AlbadaMentor: Prof. Dick van Albada
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 3
Lecture PlanLecture Plan
I.I. Context & HistoryContext & HistoryII.II. Self-reproducing loops, the evoloopSelf-reproducing loops, the evoloopIII.III. A closer lookA closer look
a)a) New method of analysisNew method of analysisb)b) Genetic, phenotypic diversityGenetic, phenotypic diversity
IV.IV. New discoveriesNew discoveriesa)a) Mutation-insensitive regionsMutation-insensitive regionsb)b) Emergent selection, cyclic genealogyEmergent selection, cyclic genealogyc)c) The evoloop as quasi-speciesThe evoloop as quasi-species
V.V. ConclusionsConclusions
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 4
ContextContext
Artificial Life:Artificial Life: Study of ”life-as-it-could-be” (Langton).Study of ”life-as-it-could-be” (Langton). Emphasizes “bottom-up” approach:Emphasizes “bottom-up” approach:
synthesize using e.g. cellular automata (CA)synthesize using e.g. cellular automata (CA) study collective behaviour emerging from study collective behaviour emerging from
local interactions (complex systems)local interactions (complex systems)
Artificial self-reproduction:Artificial self-reproduction: ““abstract from the natural self-abstract from the natural self-
reproduction problem its logical form” reproduction problem its logical form” (von Neumann)(von Neumann)
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 5
A brief historyA brief history
John von NeumannJohn von Neumann
Conway’sConway’sGame of LifeGame of Life
1950s1950s
19701970 19841984
Langton’sLangton’sSR LoopSR Loop
First international First international conference onconference onArtificial LifeArtificial Life
19891989
Chou & ReggiaChou & Reggia(emergence of replicators)(emergence of replicators)
SayamaSayama(SDSR Loop, Evoloop)(SDSR Loop, Evoloop)
19961996
Morita & ImaiMorita & Imai(shape-encoding worms)(shape-encoding worms)
Suzuki & IkegamiSuzuki & Ikegami(interaction-based(interaction-based
evolution)evolution)
20032003
Imai, Hori, MoritaImai, Hori, Morita(3D self-reproduction)(3D self-reproduction)
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 6
Self-reproduction in BiologySelf-reproduction in Biology
Traditionally (pre-1950):Traditionally (pre-1950): Self-reproduction associated with biological Self-reproduction associated with biological
systems of carbon-based organisms.systems of carbon-based organisms. Research limited by variety of natural self-Research limited by variety of natural self-
replicators.replicators. Problem of machine self-replication discussed Problem of machine self-replication discussed
purely in philosophical terms.purely in philosophical terms.
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 7
Theory of self-reproductionTheory of self-reproduction
John von Neumann (1950s):John von Neumann (1950s): First attempt to First attempt to formalizeformalize self- self-
reproduction:reproduction: Theory of Self-Reproducing AutomataTheory of Self-Reproducing Automata Universal Constructor (UC)Universal Constructor (UC)
Cellular Automata (CA) introduced Cellular Automata (CA) introduced (with S. Ulam).(with S. Ulam).
This seminal work later spawns the This seminal work later spawns the field of Artificial Life (late 1980s).field of Artificial Life (late 1980s).
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 8
The Universal The Universal ConstructorConstructor
Universal Constructor Universal Constructor (1950s):(1950s): 29 state 5-neighbour 29 state 5-neighbour
cellular automaton.cellular automaton. Capable of universal Capable of universal
construction.construction. Predicts separation between Predicts separation between
genetic information and genetic information and translators/transcribers translators/transcribers prior to discovery of prior to discovery of DNA/RNA.DNA/RNA.
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 9
Separation for evolutionSeparation for evolution
Separation is necessary for evolution:Separation is necessary for evolution: Self-description enables exact duplication.Self-description enables exact duplication. Modified self-description (by noise, etc.) Modified self-description (by noise, etc.)
introduces inexact duplication (mutation).introduces inexact duplication (mutation).
P = P = r-b-r-yr-b-r-y
CC = r-b-y-y = r-b-y-y
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 10
UC-based replication: UC-based replication: LoopsLoops
Loop structure used to represent a Loop structure used to represent a cyclic set of instructions.cyclic set of instructions. Langton (SR Loop), Morita & Imai, Chou & Langton (SR Loop), Morita & Imai, Chou &
Reggia, Sayama, Sipper, Suzuki & IkegamiReggia, Sayama, Sipper, Suzuki & Ikegami Self-replication mechanism dependent Self-replication mechanism dependent
on structural configuration of self-on structural configuration of self-replicator.replicator.
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 11
The self-reproducing The self-reproducing looploop
Sheath: Outer shell housing gene sequence.Sheath: Outer shell housing gene sequence. Genes: 7s (straight growth) and 4s (turning).Genes: 7s (straight growth) and 4s (turning). Tube: core (1) states within sheath.Tube: core (1) states within sheath. Arm: extensible loop structure for Arm: extensible loop structure for
replication.replication.
sheathsheatharmarm
tubetube
genesgenes
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 12
The evolving SR loop The evolving SR loop (evoloop)(evoloop)
A new self-reproducing loop by Sayama A new self-reproducing loop by Sayama (1999), based on SR Loop (Langton, 1984):(1999), based on SR Loop (Langton, 1984): 9-state cellular automaton.9-state cellular automaton. 5-state (von Neumann) neighbourhood.5-state (von Neumann) neighbourhood.
Modifications to earlier models (SR, SDSR) Modifications to earlier models (SR, SDSR) enable adaptivity leading to evolution.enable adaptivity leading to evolution.
Mutation mechanisms are Mutation mechanisms are emergentemergent..
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 13
Evolutionary dynamicsEvolutionary dynamics
Continuous reproduction leads to high-Continuous reproduction leads to high-density loop populationsdensity loop populations
Evolution ends with a homogeneous, Evolution ends with a homogeneous, single-species populationsingle-species population
Evolutionary dynamics seem predictable.Evolutionary dynamics seem predictable.
8
7
6
5
4
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 14
Hidden complexity?Hidden complexity?
Emergent evolutionary dynamics Emergent evolutionary dynamics demand sophisticated analysis routines.demand sophisticated analysis routines.
Original methods use size-based Original methods use size-based identification only.identification only.
Missing structural detail:Missing structural detail: gene arrangement and spacinggene arrangement and spacing genealogical ancestrygenealogical ancestry
Computational routines highly Computational routines highly expensive.expensive.
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 15
A closer lookA closer look
Loops composed of Loops composed of phenotype phenotype andand genotypegenotype:: PhenotypePhenotype: inner and outer sheath of loop: inner and outer sheath of loop GenotypeGenotype: gene sequence within loop: gene sequence within loop
Define loop species by phenotype + genotype.Define loop species by phenotype + genotype. Sufficient information for loop reconstruction.Sufficient information for loop reconstruction.
phenotypephenotypew
l
genotypegenotype
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 16
Parallels to biologyParallels to biology
The evoloop is a “messy” system:The evoloop is a “messy” system: replication is performed explicitlyreplication is performed explicitly mutation operator is emergentmutation operator is emergent interactions (collisions) produce “remnants” of inert interactions (collisions) produce “remnants” of inert
sheath states and anomalous dynamic structuressheath states and anomalous dynamic structures Birth and death must be externally defined.Birth and death must be externally defined.
remnantsremnantsdynamicdynamic
structuresstructures
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 17
Birth detectionBirth detection
Umbilical CordUmbilical CordDissolver (6)Dissolver (6)
phenotypephenotypew
l
genotypegenotype
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 18
Scan-layer trackingScan-layer tracking
Loop LayerLoop Layer
Scan LayerScan Layer
“footprint”
to parent loopto parent loop
umbilical cord dissolverumbilical cord dissolver
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 19
Death detectionDeath detection
Dissolver state
Scan layer I.D.
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 20
Labeling schemeLabeling scheme
G T C
growth turning core
G G G G C G C G T T G CC CC G
GGGGCGCGTTGCCCCG
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 21
How many permutations?How many permutations?
Constraints for exact (stable) self-replicators:Constraints for exact (stable) self-replicators: 2 2 TT-genes, -genes, nn GG-genes, (-genes, (nn-2) -2) CC-genes.-genes. TT-genes must have no -genes must have no GG-genes between them.-genes between them. Second Second TT-gene directly followed by -gene directly followed by G-G-gene.gene.
‘‘TG’TG’‘‘T’T’
n((nn-2) free ‘C’s-2) free ‘C’s
((nn-1) free ‘G’s-1) free ‘G’s
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 22
Genetic state-spaceGenetic state-space
For a loop of size For a loop of size nn, there are, there are different gene permutations resulting in exact different gene permutations resulting in exact self-replicators (stable species). self-replicators (stable species).
Do gene these permutations affect behaviour?Do gene these permutations affect behaviour?
(2n-2)(2n-2)n-2n-2( )
loop loop sizesize
# of # of speciesspecies
loop loop sizesize
# of # of speciesspecies
loop loop sizesize
# of # of speciesspecies
44 1515 99 11,44011,440 1414 9,657,7009,657,70055 5656 1010 43,75843,758 1515 37,442,16037,442,16066 210210 1111 167,960167,960 1616 145,422,67145,422,67
5577 792792 1212 646,646646,646 1717 565,722,72565,722,72
0088 3,0033,003 1313 2,496,1442,496,144 1818 2,203,961,42,203,961,4
3030
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 23
Phenotypic diversityPhenotypic diversity1000 2000 3000 4000
GCCCCGGGTTGG
GGGCGTTGCGCC
GGGGTTGCCCCG
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 24
Population dynamicsPopulation dynamics
GCCCCGGGTTGG
GGGCGTTGCGCC
GGGGTTGCCCCG
sizesize Gene sequenceGene sequence66 GGCCCCCCCCGGGGGGTTTTGGGG
77 GGCCCCGGGGGGCCGGTTTTGGCCCCGG
66 GGCCCCGGGGGGTTTTGGCCCCGG
55 GGGGCCGGTTTTGGCCCCGG
44 GGGGTTTTGGCCCCGG
44 GGGGTTTTGGCCGGCC
sizesize Gene sequenceGene sequence66 GGGGGGCCGGTTTTGGCCGGCCCC
44 GGCCGGTTTTGGCCGG
55 GGCCGGCCGGTTTTGGCCG G
sizesize Gene sequenceGene sequence66 GGGGGGGGTTTTGGCCCCCCCCGG55 GGGGGGTTTTGGCCCCCCGG44 GGGGTTTTGGCCGGCC55 GGGGCCGGTTTTGGCCGGCC44 GGGGTTTTGGCCCCG G
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 25
Emergent mutationEmergent mutation
GCCCCGGGTTGG GCCCCGGGTTGGGCCCCGGGTTGGGCCCC…
GTTGGGCCCCGGGCC GTTGGGCCCCGGGCGTTGGGCCCCG…
GGGCGTTGGGCC GGGCGTTGGGCCGGGCGTTGGGCCGGGCG…
GGCCGGGCGTTGCCCCGGCCGGGCGTTGCCGGCCGGGCGTTGCCG…
GCCGGGCGTTGCCG
(a)
(b)
(c)
(d)
(a)
(b)
(c)
(d)
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 26
Fitness landscapeFitness landscape
Evolution to both smaller Evolution to both smaller andand larger larger loops occurs.loops occurs.
Smaller loops dominate:Smaller loops dominate: higher reproductive ratehigher reproductive rate structurally robuststructurally robust
Fitness landscape balances size-Fitness landscape balances size-based fitness with genealogical based fitness with genealogical connectivity.connectivity.
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 27
Graph-based genealogyGraph-based genealogyL
oop
Si z
e
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 28
Mutation insensitive Mutation insensitive regionsregions
Certain gene subsequences are insensitive to Certain gene subsequences are insensitive to mutations:mutations:
GG{{CC}}TT{{CC}}TTGG These subsequences force a minimum loop These subsequences force a minimum loop
size.size. Evolution confined to non-overlapping subsets Evolution confined to non-overlapping subsets
of genealogy state-space.of genealogy state-space.
GGGGGGGGCCGGC C GGCCCCTTCCCCTTG GG G
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 29
New discoveriesNew discoveries
Long-term genetic diversity:Long-term genetic diversity: System continues to evolve over millions System continues to evolve over millions
of iterations.of iterations. Selection criteria not exclusively size-Selection criteria not exclusively size-
based for species with long subsequences.based for species with long subsequences. Complex evolutionary dynamics:Complex evolutionary dynamics:
Strong graph-based genealogy.Strong graph-based genealogy. Genealogical connectivity plays more Genealogical connectivity plays more
important role in selection.important role in selection.
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 30
Convergence to minimal Convergence to minimal looploop
SizeSize Gene sequenceGene sequence1414 GGGGGGGGCCGGGGGGG GGGGGGGG GTTCCCCCCCCCCCCCCCCCCCCCCTTG GG G1515 GGGGGGGGGGCCGGGGGGG GGGGGGGG GTTCCCCCCCCCCCCCCCCCCCCCCTTG G CCGG1616 GGGGGGGGGGGGCCGGGGGGG GGGGGGGG GTTCCCCCCCCCCCCCCCCCCCCCCTTG G CCCCGG1717 GGGGGGGGGGGGGGCCGGGGGGG GGGGGGGG GTTCCCCCCCCCCCCCCCCCCCCCCTTG G CCCCCCGG1515 GGGGGGGGCCGGGGGGGGGGGGGGGGC C GGTTCCCCCCCCCCCCCCCCCCCCCCTTG GG G1414 GGGGGGGGGGGGGGGGCCGGG GGGG GTTCCCCCCCCCCCCCCCCCCCCCCTTG GG G1515 GGGGGGGGGGGGGGGGCCGGGGGGGGC C GGTTCCCCCCCCCCCCCCCCCCCCCCTTG GG G1313 GGGGGGGGGG GGGGGGGGGGG GTTCCCCCCCCCCCCCCCCCCCCCCTTG GG G
11 22 33 44 55 66
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 31
Cyclic genealogyCyclic genealogy
SizeSize Gene sequenceGene sequence1818 GGGGGGGGGGGGGGG GGGGGGGGGGGGGGGG GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG GG G1919 GGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGCC G GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG GG G1919 GGGGGGGGGGGGGGGG GGGGGGGGGGGGGGGGG GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG G CCGG2020 GGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGCC G GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG G CCGG2020 GGGGGGGGGGGGGGGGG GGGGGGGGGGGGGGGGGG GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG G CCCCGG2020 GGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGCCGGCC G GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG GG G2020 GGGGGGGGGGGGGGGGG GGGGGGGGGGGGGGGGGG GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG G CCGGCC1919 GGGGGGGGGGGGGGGG GGGGGGGGGGGGGGGGG GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG GG GCC2020 GGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGGCC G GCCCCCCTTCCCCCCCCCCCCCCCCCCCCCCCCCCTTG GG GCC
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 32
ObservationsObservations
Fitness landscape:Fitness landscape: fitness fitness reproduction rate reproduction rate genealogical connectivity (cycles) genealogical connectivity (cycles) self-generated environments self-generated environments
(remnants) ?(remnants) ? Stable state is reached with Stable state is reached with
dominant species + nearest relatives.dominant species + nearest relatives. Similar to “quasi-species” model of Similar to “quasi-species” model of
Eigen, McCaskill & Schuster (1988).Eigen, McCaskill & Schuster (1988).
Section Computational Science, Universiteit van AmsterdamSection Computational Science, Universiteit van Amsterdam
University of Electro-Communications, JapanUniversity of Electro-Communications, Japan 33
ConclusionsConclusions
Simple models may hide their complexity:Simple models may hide their complexity: graph-based genealogygraph-based genealogy mutation-insensitive regionsmutation-insensitive regions emergent selection (self-generated env.)emergent selection (self-generated env.)
Sophisticated observation and Sophisticated observation and interpretation techniques play critical interpretation techniques play critical role.role.
Complex evolutionary phenomena need Complex evolutionary phenomena need not require a complex model.not require a complex model.