Maximum Likelihood Phylogenetic Inference

29 concepts29 papers1 proposal

When we compare the genomes of different organisms, we are looking at the endpoints of millions of years of independent mutations. Reconstructing the exact historical tree that connects these modern sequences is one of the fundamental challenges in evolutionary biology.

Maximum Likelihood (ML) phylogenetic inference solves this problem by framing evolution as a stochastic process. Instead of asking which tree is the most visually elegant or simplest, it asks a precise statistical question: Given a specific mathematical model of how DNA mutates, which tree topology and branch lengths make our observed sequence data most likely to have occurred?

The Core Mechanism: Modeling Sequence Substitution

To calculate the likelihood of a tree, we must first accept that mutations do not happen completely at random. For instance, transitions (mutations between purines, A $\leftrightarrow$ G, or pyrimidines, C $\leftrightarrow$ T) occur far more frequently than transversions (mutations between a purine and a pyrimidine).

As Li et al. highlight, ML methods rely on a specific model of nucleotide substitution to calculate the exact probability of these changes over time. Without this underlying model, a tree-building algorithm would treat all genetic differences identically, ignoring the biological realities of molecular evolution.

[Ancestral Sequence] 
       │
       ├───────────────► Lineage A (Accumulates transitions quickly)
       │
       └───────────────► Lineage B (Accumulates transversions slowly)

By defining these substitution rates mathematically, we can calculate the probability of a sequence transitioning from an ancestral state to a modern state across a given branch length.

Evaluating a Tree via the Likelihood Function

Imagine we have an alignment of genetic sequences from several species. To find the best evolutionary history, specialized software packages—such as IQ-TREE (utilized by Niu et al. and Wang et al.), RAxML (used by Modha et al.), or FastTree (used by Bao et al.)—evaluate a candidate tree topology.

  1. Site-by-Site Calculation: For a single column (site) in the sequence alignment, the algorithm considers all possible ancestral nucleotides at the internal nodes of the tree. Using the substitution model, it calculates the probability of those ancestors mutating into the modern nucleotides observed at the tips.
  2. Summing Ancestral States: Because we cannot observe the true ancestral sequences, the algorithm sums the probabilities across all possible combinations of internal nucleotides.
  3. Multiplying Across the Genome: Assuming each site evolves independently, the total likelihood of the entire tree is the product of the likelihoods calculated for every individual site in the alignment.

Because multiplying thousands of tiny probabilities yields numbers too small for computers to process efficiently, algorithms instead sum the natural logarithms of these probabilities, maximizing the log-likelihood score. The software uses heuristic searches to navigate the astronomical space of possible tree shapes, continuously adjusting branch lengths and swapping branches until it finds the configuration that yields the highest score.

Why Maximum Likelihood Handles Complex Evolution Better

Alternative tree-building methods often struggle when evolution behaves irregularly. For example, Maximum Parsimony operates on the principle of simplicity, favoring the tree that requires the absolute fewest mutations. However, evolution is rarely uniform.

When genes mutate at highly uneven rates across different lineages, simpler methods fall victim to artifacts like "long-branch attraction," where rapidly evolving sequences are erroneously grouped together regardless of their true relationships. As Meng et al. demonstrate, accelerated and unbalanced evolutionary rates can severely perturb simpler gene tree topologies, leading to incorrect historical inferences.

Maximum Likelihood avoids this trap because its branch lengths explicitly represent the expected number of substitutions per site. If a lineage has undergone a rapid burst of evolution, the ML framework accommodates this by lengthening that specific branch within the probability calculations, preventing the accelerated mutation rate from distorting the overall tree topology.

Measuring Certainty with the Bootstrap

Because ML inference provides the most likely tree based on a statistical model, it does not automatically guarantee that the inferred tree is the absolute truth. To quantify how robust each branch of the tree is, researchers rely on bootstrap analysis.

As emphasized by Maddison et al. and Dong et al., likelihood bootstrap analysis evaluates statistical confidence by randomly resampling columns from the original sequence alignment with replacement to generate hundreds of pseudo-alignments. The ML algorithm then reconstructs a new tree for each pseudo-alignment.

Original Alignment:  [Site 1] [Site 2] [Site 3] [Site 4]
Bootstrap Sample 1:  [Site 2] [Site 2] [Site 4] [Site 1]  ──► Reconstruct ML Tree
Bootstrap Sample 2:  [Site 3] [Site 1] [Site 1] [Site 4]  ──► Reconstruct ML Tree

The final bootstrap support value for a given branch is the percentage of these replicate trees that contain that exact evolutionary split. If a branch appears in 95% of the bootstrap trees, we can be highly confident that the signal for that relationship is distributed broadly across the genetic data, rather than being driven by a few anomalous positions.

Establishing Direction with Outgroups

By default, the mathematical equations behind ML inference calculate the relationships between sequences without determining which nodes are ancestral or descendant, resulting in an unrooted tree. To transform this network of relationships into a chronological path, researchers introduce an "outgroup."

An outgroup is a taxon known from independent biological evidence to have diverged before the lineages of interest (the ingroup). As shown in the workflows of Niu et al. and Xie et al., attaching the outgroup provides a fixed point that roots the tree. This establishes a clear directional trajectory, allowing scientists to trace traits from the common ancestor down to modern lineages.

Key Foundational Work

Phylogenetic analysis of multidrug-resistant Escherichia coli clones isolated from humans and poultry. (2013) — contributed 1 concept elements to this direction.
The deep phylogeny of jumping spiders (Araneae, Salticidae) (2014) — contributed 1 concept elements to this direction.
History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_maximum_likelihood_phylogenetic_inferenc_ec2e333c,
  title = {maximum likelihood phylogenetic inference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/maximum-likelihood-phylogenetic-inference/rdr_ec2e333cc4f64dc49b3ebefc015ee2c3},
  note = {Latest version of a living research direction; cite a version for a fixed text. Accessed 2026-11-20},
}

This exact version (version 1, frozen text):

@misc{lacuna_maximum_likelihood_phylogenetic_inferenc_ec2e333c_v1,
  title = {maximum likelihood phylogenetic inference},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/maximum-likelihood-phylogenetic-inference/rdr_ec2e333cc4f64dc49b3ebefc015ee2c3/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}
About

These directions overlap with this one through shared papers.

See something wrong? Report a mistake