Phylogenetic Inference Methods

A family of computational methods, distance, parsimony, maximum likelihood, and Bayesian, estimates evolutionary trees from molecular and morphological data.

Definition

Phylogenetic inference methods are algorithms and statistical frameworks that estimate the branching relationships among taxa from observed character data, typically under explicit models of evolutionary change.

Scope

This topic covers the major classes of tree-estimation methods: distance methods such as neighbor-joining, character-based parsimony, model-based maximum likelihood, and Bayesian inference using Markov chain Monte Carlo, along with the substitution models, optimality criteria, and software that implement them.

Core questions

• What are the main classes of tree-inference methods?
• How do distance, parsimony, likelihood, and Bayesian approaches differ?
• What role do substitution models play in inference?
• How do methods scale to large datasets?

Key theories

Maximum likelihood inference

Maximum likelihood selects the tree and model parameters that make the observed sequences most probable under an explicit substitution model, providing a statistically consistent framework.

Distance methods

Distance approaches such as neighbor-joining convert pairwise sequence differences into a tree quickly, offering speed at the cost of discarding character-level information.

Bayesian inference with MCMC

Bayesian methods sample trees in proportion to their posterior probability using Markov chain Monte Carlo, yielding both a tree estimate and a measure of uncertainty.

Clinical relevance

These methods are used to reconstruct pathogen transmission histories, date divergence events, and place newly discovered organisms, directly supporting molecular epidemiology and comparative genomics.

History

Felsenstein's likelihood framework in 1981 and Saitou and Nei's neighbor-joining in 1987 established the statistical and distance traditions; widely adopted software such as MrBayes and RAxML in the 2000s made Bayesian and large-scale likelihood analyses routine.

Debates

Speed versus accuracy across methods

Distance and parsimony methods are fast but make stronger simplifications, while likelihood and Bayesian methods are more accurate yet computationally demanding, a trade-off that shapes method choice for large datasets.

Key figures

• Joseph Felsenstein
• Masatoshi Nei
• John Huelsenbeck

Seminal works

• felsenstein1981
• saitounei1987
• ronquist2003
• stamatakis2006

Frequently asked questions

What is the difference between maximum likelihood and Bayesian phylogenetics?

Maximum likelihood finds the single tree and parameters that best fit the data, while Bayesian inference produces a distribution of trees weighted by their posterior probability, naturally expressing uncertainty.

Why are substitution models needed?

Because observed sequence differences underestimate the true number of changes when sites mutate more than once; models correct for unequal rates and multiple hits to estimate trees accurately.