bayesian phylogenetics. bayes theorem pr(tree|data) = pr(data|tree) x pr(tree) pr(data)
TRANSCRIPT
![Page 1: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/1.jpg)
Bayesian Phylogenetics
![Page 2: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/2.jpg)
Bayes Theorem
• Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree)
Pr(Data)
![Page 3: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/3.jpg)
Bayes Theorem
• Pr(Tree) = Prior probability of the tree• Pr(Data) = Prior probability of the data
– Pr(Data|Tree) over all trees, weighted by their prior probabilities
• Pr(Data|Tree) = Likelihood of the data given the tree• Pr(Tree|Data) = Posterior probability of the tree
• Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree)
Pr(Data)
![Page 4: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/4.jpg)
Assuming the prior probability of two trees (i.e., tree topologies) equal, the ratio of their posterior probabilities equals the ratio of their likelihood
scores. True or false?
Pr(Tree1|Data) = Pr(Data|Tree1) x Pr(Tree1)
Pr(Data)
Pr(Tree2|Data) = Pr(Data|Tree2) x Pr(Tree2)
Pr(Data)
Pr(Tree1|Data)
Pr(Tree2|Data)=
Pr(Data)
Pr(Data|Tree2) x Pr(Tree2)Pr(Data|Tree1) x Pr(Tree1)
Pr(Data)( )( )
![Page 5: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/5.jpg)
Assuming the prior probability of two trees (i.e., tree topologies) equal, the ratio of their posterior probabilities equals the ratio of their likelihood
scores. True or false?
Pr(Tree1|Data) = Pr(Data|Tree1) x Pr(Tree1)
Pr(Data)
Pr(Tree2|Data) = Pr(Data|Tree2) x Pr(Tree2)
Pr(Data)
Pr(Tree1|Data)
Pr(Tree2|Data)=
Pr(Data)
Pr(Data|Tree2) x Pr(Tree2)Pr(Data|Tree1) x Pr(Tree1)
Pr(Data)( )( )
![Page 6: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/6.jpg)
Assuming the prior probability of two trees (i.e., tree topologies) equal, the ratio of their posterior probabilities equals the ratio of their likelihood
scores. True or false?
Pr(Tree1|Data) = Pr(Data|Tree1) x Pr(Tree1)
Pr(Data)
Pr(Tree2|Data) = Pr(Data|Tree2) x Pr(Tree2)
Pr(Data)
Pr(Tree1|Data)
Pr(Tree2|Data)=
Pr(Data)
Pr(Data|Tree2) x Pr(Tree2)Pr(Data|Tree1) x Pr(Tree1)
Pr(Data)( )( )
![Page 7: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/7.jpg)
Assuming the prior probability of two trees (i.e., tree topologies) equal, the ratio of their posterior probabilities equals the ratio of their likelihood
scores. True or false?
Pr(Tree1|Data) = Pr(Data|Tree1) x Pr(Tree1)
Pr(Data)
Pr(Tree2|Data) = Pr(Data|Tree2) x Pr(Tree2)
Pr(Data)
Pr(Tree1|Data)
Pr(Tree2|Data)= Pr(Data|Tree1)
Pr(Data|Tree2)
![Page 8: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/8.jpg)
If tree topology is the parameter of interest, what are some “nuisance parameters” that need to be accommodated in a Bayesian or Maximum
Likelihood analysis?
• Branch lengths
• Substitution rates
• Base frequencies
• Rate heterogeneity parameters
![Page 9: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/9.jpg)
How do Bayesian and Likelihood approaches differ in their treatment of nuisance
parameters?
• Likelihood– Find the value of each parameter that
maximizes the likelihood of the data
• Bayesian– Integrate over all possible values of each
parameter (weighted by the prior probability distribution)
![Page 10: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/10.jpg)
How does MCMC get around the problem of not being able to calculate Pr(Data)?
• The probability of staying/leaving a tree is determined by its posterior probability relative to other nearby trees
• The overall time spent on a tree will converge to its absolute posterior probability
![Page 11: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/11.jpg)
Like walking-over tree space at a rate governed by altitude..
![Page 12: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/12.jpg)
What happens at each “step” of a Bayesian, phylogenetic MCMC analysis?
• A new parameter (topology, branch lengths, substitution parameters etc.) is proposed
• Whether the new parameter is accepted is governed by the metropolis-hasting equation
Pr(Data|x= j)
Pr(Data|x = i
Pr(x = j)
Pr(x = i
Pr(proposing x=i| x=j)
Pr(proposing x=j| x=i)x x1,min
Posterior ratio Prior ratio Proposal ratio
![Page 13: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/13.jpg)
Why is the output of MCMC called a “posterior distribution?” What does it
contain?
• It contains a list of parameters in effect at a sampled set of generations (after burnin)
• The frequency of a parameter in this sample should be proportional to its posterior probability
![Page 14: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/14.jpg)
How can the posterior distribution be queried to find the posterior probability of a clade?
• Just see the proportion of trees in the distribution that have the clade
![Page 15: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/15.jpg)
How can the posterior distribution be queried to evaluate other parameters, for example the
transition:transversions ratio?
• Generate a histogram (or fit to a probability density function)
• Establish a credibility interval – the range than encompasses, say, 95% of the distribution
![Page 16: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/16.jpg)
A dice example• A manufacturer makes regular dice and trick dice (with 2
sixes) – in equal numbers• You are given a die from this manufacturer and are not
allowed to look at all the sides – you can only look at the side that is up after a roll
• You roll this die 10 times and get these numbers
• What is the probability that the die is one with two sixes?
![Page 17: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/17.jpg)
Bayesian approach
Pr(H|D) = Pr(D|H) x Pr(H) Pr(D)
1-six 2-six Sum
Pr(H)
Pr(D|H)
![Page 18: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/18.jpg)
Bayesian approach
Pr(H|D) = Pr(D|H) x Pr(H) Pr(D)
1-six 2-six Sum
Pr(H) 1.0
Pr(D|H)
![Page 19: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/19.jpg)
Bayesian approach
Pr(H|D) = Pr(D|H) x Pr(H) Pr(D)
1-six 2-six Sum
Pr(H) 0.5 0.5 1.0
Pr(D|H)
![Page 20: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/20.jpg)
Bayesian approach
Pr(H|D) = Pr(D|H) x Pr(H) Pr(D)
1-six 2-six Sum
Pr(H) 0.5 0.5 1.0
Pr(D|H) (1/6)5*(5/6)3 (2/6)5*(4/6)3
![Page 21: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/21.jpg)
Bayesian approach
Pr(H|D) = Pr(D|H) x Pr(H) Pr(D)
1-six 2-six Sum
Pr(H) 0.5 0.5 1.0
Pr(D|H) 7.44E-05 0.00123 0.00129
![Page 22: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/22.jpg)
Bayesian approachPr(H|D) = 0.00123 x 0.5 = 0.953
0.00129 x 0.5
1-six 2-six Sum
Pr(H) 0.5 0.5 1.0
Pr(D|H) 7.44E-05 0.00123 0.00129
Pr(H|D) = 0.00006 x 0.5 = 0.047 0.00129 x 0.5
![Page 23: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/23.jpg)
MCMC approach
1 six 2 sixes
Pr(Data|x= j)
Pr(Data|x = i
Pr(x = j)
Pr(x = i
Pr(proposing x=i| x=j)
Pr(proposing x=j| x=i)x x1,min 1 1
![Page 24: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/24.jpg)
Pr(Data|one six)
Pr(Data|two sixes
MCMC approach
1 six 2 sixes
1,min = 1
![Page 25: Bayesian Phylogenetics. Bayes Theorem Pr(Tree|Data) = Pr(Data|Tree) x Pr(Tree) Pr(Data)](https://reader035.vdocuments.mx/reader035/viewer/2022070411/56649f4e5503460f94c7047e/html5/thumbnails/25.jpg)
Pr(Data|two sixes)
Pr(Data|one six
MCMC approach
1 six 2 sixes
1,min = Pr(Data|two sixes)
Pr(Data|one six = 0.047
If you run this long enough, you will spend only 4.7% of the time on “1 six.” This is the PP of that hypothesis.