This talk introduces the use of experimental phylogenetics via digital evolution as being a complementary tool to other methods that investigate phylogenetic accuracy. Unlike the experimental generation of evolutionary histories using biological organisms, digital evolution is immensely more feasible, allows a large range of experimental treatment investigation, and maintains a perfect record of all information regarding the evolutionary history. Such information includes all genomic sequences at each generation, which allows for an empirical model of substitution, a detailed characterization of homoplasious characters, etc. Unlike computational simulation, digital evolution maintains a greater degree of biological realism. There are many elements important to the evolving system that are not set a priori by the investigator. For example, since Avida exhibits natural selection inherently (rather than being simulated) there is a distribution of mutational effects and very complex epistatic interactions that change dynamically along with the evolving system. In addition, population processes such as genetic drift, hitchhiking, etc are all innate components. The results presented here are a mere scratch on the surface of possibilities. The application of digital experimental evolution to phylogenetic methodological design and evaluation has the potential to be substantial. For questions, ideas, collaborations, etc please email kohncory@msu.edu

1. A New Age of Experimental Phylogenetics:
Digital Evolution and the Population Processes that
Reduce Phylogenetic Accuracy
Cory Kohn
Barry L. Williams
Dept. of Zoology, EEBB Graduate Program,
BEACON Center for the Study of Evolution In Action
Michigan State University

2. How do we determine whether
phylogenetic methods are accurate?
• Mathematical and statistical analyses
• Computer simulation
• Comparison with known phylogenies
– Experimental or otherwise

3. Simulation approaches have several
drawbacks
• Explicit simplifications using proposed models
• Complex or emergent properties not easily
simulated
– Natural selection
– Epistatic interactions
– Hill-Robertson or clonal interference
– Genetic hitchhiking, etc

4. Analyses of known histories also have
drawbacks
• Difficult and costly
• Lack of replication
• Evolutionarily limited change
• Not representative of nature
– Provides information on one actual history…
Rather than “exhaustive information on an idealized
set of conditions that never actually exist in nature”
Hillis, DM et al. (1993), Experimental approaches to phylogenetic analysis. Systematic Biology 42:90-92.

5. Experimental phylogenetics:
Landmark study with T7 phage
• Single clone lineage seed
• Symmetrical, periodic
cladogenesis
Hillis, DM et al. (1992), Experimental phylogenetics: generation of a known phylogeny. Science 255:589-592.
Hillis, DM et al. (1994), Application and accuracy of molecular phylogenies. Science 264:671-677.

6. Experimental phylogenetics:
Landmark study with T7 phage
• Single clone lineage seed
• Symmetrical, periodic
cladogenesis
• RE data: all methods
inferred the true tree
• Seq data: ML inferred 5
of 6 clades correctly
Hillis, DM et al. (1992), Experimental phylogenetics: generation of a known phylogeny. Science 255:589-592.
Hillis, DM et al. (1994), Application and accuracy of molecular phylogenies. Science 264:671-677.

7. Requirements for phylogenetics
• Heritable traits with variation
• Ancestor – descendant relationships
• Series of bifurcations (cladogenesis)
• Hypothesized model of evolution

8. Requirements for phylogenetics
• Heritable traits with variation
• Ancestor – descendant relationships
• Series of bifurcations (cladogenesis)
• Hypothesized model of evolution

9. Avida is a software platform for
evolution in silico
Population
of digital organisms

10. A digital organism is a small computer
program
Population
of digital organisms

11. The genome consists of a sequence of
computational instructions
Population
of digital organisms

12. Self-replication of
instruction-set
genome with
probabilistic
mutation rate
The genome allows self-replication
Population
of digital organisms

13. Population
of digital organisms
Self-replication of
instruction-set
genome with
probabilistic
mutation rate
Phenotype provides
reward depending
on environment
The genome allows self-replication
and the ability to perform tasks

14. Self-replication of
instruction-set
genome with
probabilistic
mutation rate
The genome allows self-replication
and the ability to perform tasks
Heritability
Population
of digital organisms
Phenotype provides
reward depending
on environment

15. The genome allows self-replication
and the ability to perform tasks
Heritability
Variation
Population
of digital organisms
Phenotype provides
reward depending
on environment

16. The genome allows self-replication
and the ability to perform tasks
Heritability
Variation
Differential
Fitness
Population
of digital organisms

17. Adaptive evolution, genetic drift,
population level processes can occur
Heritability
Variation
Differential
Fitness
Adaptive Evolution
Population
of digital organisms

18. Avida sequence data
resembles an amino acid sequence
• Avida genome:
utycasvabrucavcqgfcqapqcccccccc

19. Avida sequence data
resembles an amino acid sequencealignment

20. • Known mutational model (Poisson)
Avida sequence data
has a known model of evolution

21. • Known mutational model (Poisson)
• Knowable substitution model
Avida sequence data
has a known model of evolution

22. Requirements for phylogenetics
• Heritable traits with variation
• Ancestor – descendant relationships
• Series of bifurcations (cladogenesis)
• Hypothesized model of evolution

23. Artificial life is an attractive alternative
Heritable traits with variation
Ancestor – descendant relationships
Series of bifurcations (cladogenesis)
• Hypothesized model of evolution
We can do even better!

24. Artificial life is an attractive alternative
Heritable traits with variation
Ancestor – descendant relationships
Series of bifurcations (cladogenesis)
• Hypothesized model of evolution
We can do even better!
Easy and cheap
Evolutionary relevant change
Powerfully replicable

25. Do experimental histories generated
with artificial life resolve as expected
under simplistic conditions?

26. Do experimental histories generated
with artificial life resolve as expected
under simplistic conditions?
Can results inform biological reality?

27. Replicated T7 study using Avida
• Approximately replicated
substitution rate
• Perfect alignment
• Poisson model of
molecular evolution
• MrBayes
• RAxML

28. Extended to include various factors
• Recombination;
Lineage seeding
– Asexual; Single clone
– Recombination; Population
• Natural selection
– Population: 10 or 1000
• Extent of evolution
– 100, 300, or 3000 generations per lineage
– uniform, varied

29. Experimental phylogenetics
True

30. Experimental phylogenetics
True
H qtcahclvdqvnc…
I qtcahcdnqsgrc…
J qtcalcivqdcrm…
K qtcayyivqdcrc…
Experimental data

31. Experimental phylogenetics
True Inferred
H qtcahclvdqvnc…
I qtcahcdnqsgrc…
J qtcalcivqdcrm…
K qtcayyivqdcrc…
Experimental data

32. True Inferred
Evaluation criterion: Clade Accuracy

33. True Inferred
Evaluation criterion: Clade Accuracy
Robinson-Foulds distance = 2
Normalized RF distance = 0.17

34. True Inferred
Clade Accuracy = 1 – normalized RF distance
Percent of clades correctly identified in the inferred tree
Evaluation criterion: Clade Accuracy

35. Evaluation criterion: Clade Accuracy
True
Clade Accuracy
5/6 = 83%
Inferred

36. Evaluation criterion: Clade Accuracy
True
10 replicates per
experimental
conditions
Clade Accuracy
41/60 = 68%

37. Clade Accuracy plots

38. Simple, neutral evolution

39. Strong stabilizing selection

40. Recombination
population
transfers

41. Differing extent of lineage evolution
SLL: Short, Long, Long
300, 3000, 3000 generations per branch
reading backwards in time

42. Differing extent of lineage evolution
LSSB – Branch Breakup
Add taxa to equalize
relative extent of evolution

43. Differing extent of lineage evolution
SLL
LSL
LLS
LSS

44. Long branch attraction
LSL LSS
66% 33%
Parsimony “informative”
sites
SLL
LSL
LLS
LSS

45. Varying degrees of adaptive evolution
1000 organisms
in population

46. Adaptive evolution with or without
recombination

47. Adaptive evolution with or without
recombination

48. Lineage evolution, adaptive evolution,
recombination

49. Lineage evolution, adaptive evolution,
recombination

50. Unpredicted combinatorial effects

51. Tree posterior probability
indicative of clade accuracy

52. Conclusions
• MrBayes and RAxML performed well
– Even under difficult conditions
– Tree posterior indicative of clade accuracy
• Poor performance results from combinatorial
effects of factors
• Digital evolution is an attractive companion to
simulations and biological experimental
phylogenetics

53. Acknowledgements
Williams Lab:
Kyle Safran
Carlos Anderson
Kevin Hall
David Hillis
April Wright
avida.devosoft.org
Learn
More!
beacon-center.org
Resources and Funding:

54. All treatments

55. Varying degrees of adaptive evolution
Scheme 1 Scheme 2 Scheme 3 Scheme 4
1 environment 2 environments 6 environments 14 environments

56. Unpredicted combinatorial effects

57. 20 available instructions » AA
# No-ops
● nop-A 1 # a
● nop-B 1 # b
● nop-C 1 # c
# Flow control operations
● if-n-equ 1 # d
● #if-less 1 #
● if-label 1 # e
● mov-head 1 # f
● jmp-head 1 # g
● get-head 1 # h
● #set-flow 1 #
# I/O and Sensory
● IO 1 # s
● h-search 1 # t
recoded w y v to j o b
# Single Argument Math
● #shift-r 1 #
● #shift-l 1 #
● inc 1 # i
● #dec 1 #
● push 1 # j
● pop 1 # k
● swap-stk 1 # l
● swap 1 # m
# Double Argument Math
● #add 1 #
● sub 1 # n
● nand 1 # o
# Biological Operations
● h-copy 0 # p
● h-alloc 1 # q
● h-divide 1 # r