Bayesian Phylogenetics: Priors, Sampling and Judging Convergence
A Bayesian analysis returns a distribution over trees rather than a single tree, sampled by a Markov chain. This stop covers what the priors do to the answer and how to tell a converged chain from one that only looks settled.
What a learner can do afterwards
- State what a posterior probability on a clade claims, and how it differs from a bootstrap proportion
- Explain the effect of the tree prior and the clock prior on a dated analysis
- Diagnose a chain from effective sample size, trace plots and agreement between independent runs
1 · Read
You estimate the tree with uncertainty instead of picking one best tree. Your result is a posterior distribution over trees. The posterior probability of a clade is the share of that distribution that contains the clade. A bootstrap proportion measures stability under resampling, which answers a different question from a posterior probability.
You combine the likelihood of the sequences with your priors. The tree prior encodes expectations about speciation and extinction, and it matters most for node ages. The clock prior sets how freely rates vary across branches, from one strict shared rate to relaxed models.
You date a tree from short sequences and get old nodes. You rerun with a different tree prior and the dates move a lot. That swing tells you the prior is carrying the conclusion, so you report both runs and admit the data are thin.
You check the sampler before you quote any number. You look for high effective sample sizes, stable traces without trends, and independent runs that agree on clade posteriors. If any check fails, the chain has not converged and its numbers stay private.
You read posteriors as belief given data and model, you own your priors, and you quote no number until the chain converges.
2 · Watch
Take it off screen
Where it sits
8 questions wait behind this lesson, each with its answer explained. Every answer feeds the sky: stars light as they are learned, and dim when it is time to come back.