Borges, J. L On exactitude in science. P. 325, In, Jorge Luis Borges, Collected Fictions (Trans. Hurley, H.) Penguin Books.

Transcription

1 ... In that Empire, the Art of Cartography attained such Perfection that the map of a single Province occupied the entirety of a City, and the map of the Empire, the entirety of a Province. In time, those Unconscionable Maps no longer satisfied, and the Cartographers Guilds struck a Map of the Empire whose size was that of the Empire, and which coincided point for point with it. The following Generations, who were not so fond of the Study of Cartography as their Forebears had been, saw that that vast Map was Useless, and not without some Pitilessness was it, that they delivered it up to the Inclemencies of Sun and Winters. In the Deserts of the West, still today, there are Tattered Ruins of that Map, inhabited by Animals and Beggars; in all the Land there is no other Relic of the Disciplines of Geography. Suárez Miranda, Viajes de varones prudentes, Libro IV, Cap. XLV, Lérida, 1658 Borges, J. L On exactitude in science. P. 325, In, Jorge Luis Borges, Collected Fictions (Trans. Hurley, H.) Penguin Books.

2 Fitting (more) Macroevolutionary Models to Data Luke J. Harmon

3 Slatkin and Pollack 2005

4 within species Slatkin and Pollack 2005

5 within species among species Slatkin and Pollack 2005

6 Hansen and Martins 1996

7 Topics Fitting models to comparative data: what do we know? Extending the set of models we can fit The future of comparative methods

8 Topics Fitting models to comparative data: what do we know? Extending the set of models we can fit The future of comparative methods

9 Example: Anolis lizards Lizards on Caribbean islands Phylogenetic and body size data for 73 species (out of ~140 total) Anolis baleatus

10

11 Brownian Motion Two parameters: starting value (zo) and rate (σ 2 ) dz(t) = σ db(t) zo t z(t)

12 Phylogeny A B

13 var(a) Phylogeny A B

14 var(a) Phylogeny A var(b) B

15 var(a) Phylogeny A cov(a,b) B var(b) A B

16 Phylogenetic variance-covariance (VCV) matrix A B A B

17 Phylogenetic variance-covariance (VCV) matrix var(a) A var(b) B A B

18 Phylogenetic variance-covariance (VCV) matrix var(a) cov(a,b) A cov(a,b) var(b) B A B

19 Phylogenetic variance-covariance (VCV) matrix var(a) cov(a,b) A σ 2 cov(a,b) var(b) B A B

20 genetic drift (co)variance + -

21 General form Tip data follow a multivariate normal distribution with mean vector zo and variance-covariance matrix where var(i) = σ 2 (di); di =distance from root to tip i cov(i,j) = σ 2 (ci,j); ci,j =shared path of tip i and j

22 We can fit Brownian motion model to comparative data using likelihood

23 16 17 σ 2 Sigma Squared 0e+00 1e 09 2e 09 3e 09 4e zo Theta

24 Quantitative genetics A quantitative genetics model of pure genetic drift also produces Brownian motion Three parameters: G, Ne, zo σ 2 = G/Ne

25 G G 1e 11 1e 10 1e 09 1e 08 1e 07 1e Ne Ne

26 G G 1e 11 1e 10 1e 09 1e 08 1e 07 1e σ 2 = G/Ne Ne Ne

27 σ 2 = G/Ne

28 σ 2 = G/Ne σ 2 / Vp = h 2 /Ne

29 Across a wide range of taxa, σ 2 /Vp is about 0.74

30 Across a wide range of taxa, σ 2 /Vp is about 0.74 That is, the average Brownian rate parameter is about 0.74 phenotypic standard deviations per million years

31 Across a wide range of taxa, σ 2 /Vp is about 0.74 That is, the average Brownian rate parameter is about 0.74 phenotypic standard deviations per million years That translates to a variance of 1.2 x 10-6 phenotypic sd per generation

32 Across a wide range of taxa, σ 2 /Vp is about 0.74 That is, the average Brownian rate parameter is about 0.74 phenotypic standard deviations per million years That translates to a variance of 1.2 x 10-6 phenotypic sd per generation About half of the time, the change from one generation to the next is phenotypic s.d.

33 G G 1e 11 1e 10 1e 09 1e 08 1e 07 1e σ 2 = G/Ne Ne Ne

34 G G 1e 11 1e 10 1e 09 1e 08 1e 07 1e Evolution is too slow for drift σ 2 = G/Ne Ne Ne

35 What about selection?

36 Brownian motion can also result if selection is random in direction and relatively weak

37

38 BM model: About half of the time, the change from one generation to the next is phenotypic s.d.

39 What happens with OU models?

40 OU Model - single optimim Three parameters: starting value (μ), rate (σ 2 ), and constraint parameter (α) i sij j T = total tree depth

41 stabilizing selection (co)variance + -

42 α = 1 / (ω+p)

43 α = 1 / (ω+p) mean α across clades: 0.34

44 α = 1 / (ω+p) mean α across clades: 0.34 Typical ω 2 = 3-50

45 α = 1 / (ω+p) mean α across clades: 0.34 Typical ω 2 = 3-50 P would have to be negative to get these alpha values

46 α = 1 / (ω+p) mean α across clades: 0.34 Typical ω 2 = 3-50 P would have to be negative to get these alpha values (e.g. stabilizing selection is typically stronger than what OU values suggest)

47 What does this mean? Commonly used models can fit comparative data Simplistic quantitative genetics interpretations of these models are probably not correct

48 Brownian motion is not drift Ornstein-Uhlenbeck is not stabilizing selection on a single peak

49 Hansen-Martins environmental change model Imagine populations are subject to strong stabilizing selection on optima But the position of these optima varies according to a Brownian motion model

50 σb 2 = overall rate of drift σe 2 = rate of drift of optima Hansen and Martins 1996

51 σb 2 = overall rate of drift σe 2 = rate of drift of optima When σe 2 larger than (left term): Hansen and Martins 1996

52 σb 2 = overall rate of drift σe 2 = rate of drift of optima When σe 2 larger than (left term): σb 2 = σe 2 Hansen and Martins 1996

53 Main idea: patterns of trait means on trees might reflect the dynamics of the adaptive landscape more than they reflect processes of adaptation within populations

54 Slatkin and Pollack 2005

55 within species Slatkin and Pollack 2005

56 within species among species Slatkin and Pollack 2005

57 Possible model: The location of the optimum changes rapidly from one generation to the next Globally, optima are constrained to be within a certain range of values

58 Topics Fitting models to comparative data: what do we know? Extending the set of models we can fit The future of comparative methods

59 Extending Models Solving likelihoods for new models Using Bayesian approaches ABC

60 Extending Models Solving likelihoods for new models Using Bayesian approaches ABC

61 Early Burst Model (EB) Rate of evolution slows through time Highest rate at the root of the tree Three parameters: starting value (μ), starting rate (σ 2 o), and rate change (r) i sij j

62 Proportion of weight Proportion of weight Body size Body shape Squamates Birds Fish Insects Mammals BM CC EB NA Amphibians

63 Extending Models Solving likelihoods for new models Using Bayesian approaches ABC

64 Acceptance probability Prior =. odds Likelihood ratio Standard Bayesian MCMC

65 rjmcmc reversible-jump MCMC rjmcmc is an MCMC algorithm that can jump between models of differing complexity

66 rjmcmc moves Update parameter values (root state, rate σ 2 i) Split single rate category into two Merge two rate categories into one

67 (α1 α1 α1 α1 α1 α1 α1 α1 α1 α1) k = 1

68 (α1 α1 α1 α1 α1 α1 α1 α1 α1 α1) k = 1 split move (α1 α1 α1 α1 α1 α2 α2 α2 α2 α2) k = 2

69 (α1 α1 α1 α1 α1 α1 α1 α1 α1 α1) k = 1 split move (α1 α1 α1 α1 α1 α2 α2 α2 α2 α2) k = 2 split move (α3 α3 α3 α1 α1 α2 α2 α2 α2 α2) k = 3

70 (α1 α1 α1 α1 α1 α1 α1 α1 α1 α1) k = 1 split move (α1 α1 α1 α1 α1 α2 α2 α2 α2 α2) k = 2 split move (α3 α3 α3 α1 α1 α2 α2 α2 α2 α2) k = 3 merge move (α3 α3 α3 α1 α1 α1 α1 α1 α1 α1) k = 2

71 Acceptance probability Prior =. odds Likelihood ratio Standard Bayesian MCMC

72 Acceptance probability Prior =. odds Likelihood ratio. Hastings ratio Standard Bayesian MCMC Reversible jump MCMC

73 AUTEUR ACCOMMODATING UNCERTAINTY IN TRAIT EVOLUTION USING R Jonathan M. Eastman, Michael E. Alfaro, Paul Joyce, Andrew E. Hipp, and Luke J. Harmon

74 AUTEUR Do rates of trait evolution vary across clades in a phylogenetic tree? Are traits in some clades evolving faster than in others?

75

76 Em

77 Em Graptemys Pseudemys

78 Extending Models Solving likelihoods for new models Using Bayesian approaches ABC

79 Standard Bayes MCMC Acceptance probability Prior =. odds Likelihood ratio

80

81 Prior Density θ

82 Prior Density θ

83 Prior compute likelihood of data under model M with θ Density θ

84 Prior compute likelihood of data under model M with θ Density Likelihood ratio θ

85 Prior compute likelihood of data under model M with θ accept proposal with probability h Posterior Density Likelihood ratio Density θ θ

86 Prior compute likelihood of data under model M with θ accept proposal with probability h Posterior Density Likelihood ratio Density θ θ

87 Prior compute likelihood of data under model M with θ accept proposal with probability h Posterior Density Likelihood ratio Density θ θ otherwise reject

88 Prior compute likelihood of data under model M with θ accept proposal with probability h Posterior Density Likelihood ratio Density θ θ otherwise reject

89 Standard Bayes MCMC Acceptance probability Prior =. odds Likelihood ratio

90 Standard Bayes MCMC Acceptance probability Prior =. odds Likelihood ratio But what if we can t compute the likelihood ratio?

91 Standard Bayes MCMC Acceptance probability Prior =. odds Likelihood ratio But what if we can t compute the likelihood ratio? Use Approximate Bayesian Computation (ABC)

92 ABC

93 ABC Prior Density θ

94 ABC Prior Density θ

95 ABC Prior simulate data under model M with θ Density θ

96 ABC Prior simulate data under model M with θ Density θ

97 ABC Prior simulate data under model M with θ Density θ

98 ABC Prior simulate data under model M with θ Density θ

99 ABC Prior simulate data under model M with θ Density θ

100 ABC Prior simulate data under model M with θ Density θ

101 ABC Prior simulate data under model M with θ Posterior Density Density θ θ

102 ABC Prior simulate data under model M with θ Posterior Density Density θ θ

103 MECCA Simulation Algorithm

104 n

105 1. Draw ancestral character state Θ n 3 5 Θ

106 1. Draw ancestral character state Θ n 3 5 Θ Simulate characters along branches under BM

107 3. Simulate characters in unresolved clades using a branching diffusion process (birth-death plus Brownian motion)

108 ABC summary statistics

109 ABC summary statistics

110 ABC summary statistics a

111 ABC summary statistics σ 2 a

112 ABC summary statistics σ 2 a a σ 2

113 λ θμ μ a

114 λ λ μ θμ a

115 λ λ μ θμ a

116 λ λ μ θμ a σ 2 σ 2

117 λ λ μ θμ a a σ 2 σ 2

118 λ λ μ θμ a a σ 2 σ 2

119 λ λ μ θμ a a σ 2 σ 2

120 Diversification rates

121 Diversification rates Lambda Mu density Lambda Mu Index

122 Diversification rates Lambda Mu density Lambda Mu λ = 0.19 ( ) μ = 0.10 ( ) Index

123 Character evolution Density Density σ 2 sigmasq lnθ Root State

124 Character evolution Density Density σ 2 sigmasq ( ) lnθ Root State 7.1 kg ( )

125 ACME C MECCA Modeling the Evolution of Continuous Characters using ABC Graham Slater, Luke Harmon, Paul Joyce, Liam Revell, and Michael Alfaro

126 Topics Fitting models to comparative data: what do we know? Extending the set of models we can fit The future of comparative methods

127 Fitting the data people actually have Incomplete data where sampling is nonrandom Combinations of different types of data, like within and among species

128 Adding to the set of models Moving beyond Brownian motion Estimating meaningful parameters Adding complexity when possible

129 New Statistical Tools Dealing with all forms of uncertainty More flexible Bayesian analyses ABC

130 Topics Fitting models to comparative data: what do we know? Extending the set of models we can fit The future of comparative methods