BIO 370 Lecture Notes - Lecture 25: Likelihood Function, Theria, Confidence Interval
Inferring Phylogeny
Once a tree has been inferred from the data, how certain can we be that this tree (or some
component of this tree) is correct?
Bootstrap support levels and statistical significance levels are not the same thing.
We can bootstrap data for parsimony, distance and ML methods.
Presumably it could be done for Bayesian, but there are no computer programs to do so.
ML and Bayesian methods can be taken a step further.
100 = 100% bootstrap support.
Not all branches on this tree are equally supported.
Most computer programs allow one to input a tree representing a particular
hypothesis.
• So they created a tree with a monophyletic Marsupials plus Monotremes, with Theria
as the sister group.
• They calculated the best score such a tree could have and compared that to the best
score obtained with all OTUs allowed to move about freely.
They were also able to determine that alternative hypotheses were significantly
different than this tree which had the best likelihood score.
find more resources at oneclass.com
find more resources at oneclass.com
Odds Ratio Testing
• Likelihood functions are automatically calculated in Bayesian and ML methods.
• Associated with this likelihood function for a given dataset and tree is a distribution of
the likelihood parameter.
• Like any other parameter with a known or estimated distribution, we can use this to
test whether the likelihoods that two different trees impart to the dataset are the
same or different
• (whether the two trees belong to the same confidence set or not – akin to
whether two means belong to the same confidence interval.)
Phylogeny and the Comparative Method
Why is it iportat to ilude phylogey as a effet i aalysis of ariatio i pheotypi
traits?
Because species of interest share a common evolutionary history, and historical relationships
among them are represented by a phylogeny.
What’s the potetial prole?
How many degrees of freedom do we really have?
Are the data points truly independent? How many times did these traits/habitat preferences
evolve?
Problem: chi-squared test does not take shared evolutionary history into account, assumes that
each sample is independent from each other.
9 degrees of freedom
Not truly independent because the entire pattern has arisen from a single pair of evolutionary
changes
This tree was calculated using DNA and/or morphological data, and then habitat preference
mapped onto it as we would map any trait.
find more resources at oneclass.com
find more resources at oneclass.com
Document Summary
Bootstrap support levels and statistical significance levels are not the same thing. We can bootstrap data for parsimony, distance and ml methods. Presumably it could be done for bayesian, but there are no computer programs to do so. Ml and bayesian methods can be taken a step further. Not all branches on this tree are equally supported. They were also able to determine that alternative hypotheses were significantly different than this tree which had the best likelihood score. Likelihood functions are automatically calculated in bayesian and ml methods: associated with this likelihood function for a given dataset and tree is a distribution of the likelihood parameter. Because species of interest share a common evolutionary history, and historical relationships among them are represented by a phylogeny. Problem: chi-squared test does not take shared evolutionary history into account, assumes that each sample is independent from each other.