Probabilistic Inference of Phylogenetic Trees

41 concepts41 papers1 proposal

When biologists or oncologists look at a set of genetic sequences, they see a historical record. Reconstructing the exact sequence of evolutionary splits that led to those sequences—the phylogenetic tree—is one of the foundational challenges of computational biology.

The core difficulty is scale. For a set of $N$ organisms, the number of possible unrooted tree topologies grows double-factorially: $(2N-5)!!$. For just 50 species, the number of possible trees vastly exceeds the number of atoms in the observable universe. Early methods relied on Maximum Parsimony, which simply looks for the tree that requires the fewest total mutations. However, evolution is rarely parsimonious; mutations can happen at different rates across lineages, traits can be lost, and nucleotides can mutate back to their original states.

To overcome these limitations, modern computational biology relies on probabilistic frameworks—specifically Maximum Likelihood (ML) and Bayesian Inference (BI). Rather than just counting changes, these methods model the underlying biochemistry of sequence evolution mathematically, allowing researchers to navigate the astronomical space of possible trees and quantify their uncertainty.

Maximum Likelihood: Evaluating a Single Hypothesis

Maximum Likelihood asks a straightforward statistical question: Given a specific tree topology, branch lengths, and a mathematical model of nucleotide substitution, what is the probability that we would observe our actual sequence alignment?

The algorithm treats the tree topology and branch lengths as parameters to be optimized. It computes the likelihood of the data at each site in the alignment, sums them across all sites, and adjusts the tree structure to maximize this value. This approach works because it explicitly accounts for hidden evolutionary events, such as a nucleotide changing from A to T and back to A over long periods.

In practice, researchers use highly optimized software suites to perform this search. For example, Xuemei Yang et al. aligned chloroplast genomes using MAFFT and reconstructed the evolutionary history of oak species via RAxML in The first complete chloroplast genome of Quercus coccinea (Scarlet Oak) and its phylogenetic position within Fagaceae. Similarly, Yunqi Liu et al. leveraged the IQ-TREE framework to determine the evolutionary placement of endangered plants in The complete chloroplast genome of Ziziphus mairei Dode 1908 (Rhamnaceae), an endangered perennial plant in Yunnan, China. By weighting transitions and transversions based on empirical data, ML provides a highly accurate estimate of the single "best" tree.

Bayesian Inference and MCMC: Mapping the Landscape

While Maximum Likelihood searches for a single optimal peak, Bayesian Inference shifts the paradigm to map the entire landscape. It asks: What is the probability distribution of all possible trees given our data?

This approach relies on Bayes' Theorem to calculate the posterior probability of a tree topology. However, calculating the exact denominator of Bayes' Theorem requires integrating over every single possible tree configuration—an analytical impossibility.

To solve this, researchers use Markov Chain Monte Carlo (MCMC) algorithms. Instead of calculating the entire space, an MCMC chain takes an intelligent random walk through tree space. The chain starts at a random tree and proposes a small modification (such as swapping two branches). If the new tree explains the data better (higher likelihood), the chain moves there. If it is worse, the chain might still move there with a certain mathematical probability, preventing it from getting trapped in local optima.

Over millions of generations, the chain spends more time in high-probability regions. The frequency with which a specific tree topology is visited converges to its true posterior probability.

This probabilistic sampling is widely used to track rapidly mutating pathogens and ancient lineages:

The Intuition of Multi-Method Cross-Validation

Because every statistical model introduces specific assumptions, relying on a single algorithmic approach can introduce systemic biases. For instance, Maximum Parsimony can suffer from "long-branch attraction," erroneously grouping rapidly evolving lineages together.

To ensure their biological conclusions are robust, researchers frequently employ a multi-method strategy, running their data through ML, BI, and parsimony frameworks simultaneously to look for topological consensus. If completely different mathematical formulations yield identical branching patterns, confidence in the resulting tree increases significantly.

As Chenghao Jia et al. noted in Comparative analysis of the complete mitochondrial genomes of three rockfishes (Scorpaeniformes, Sebastiscus ) and insights into the phylogenetic relationships of Sebastidae, utilizing multiple inference methods ensures consistency across different computational approaches. This cross-validation is critical when classifying organisms with highly ambiguous or convergent traits:

Scale Invariance: From Global Pandemics to Tumors

The mathematical beauty of probabilistic tree inference is its scale invariance. The exact same algorithms used to map deep evolutionary time across millions of years are regularly deployed to map rapid, real-time genetic shifts.

In epidemiology, tracking outbreaks requires calculating divergence times and evolutionary rates from contemporary sequences. Tomasz A. Łęski et al. used Bayesian MCMC to model the relationships between contemporary viral sequences and historical genomic data in Sequence Variability and Geographic Distribution of Lassa Virus, Sierra Leone, revealing hidden genetic diversity. Similarly, Massimo Ciccozzi et al. applied Bayesian MCMC to estimate the posterior probability distribution of evolutionary trees for multidrug-resistant bacteria, providing a probabilistic framework to confirm zoonotic transmission between poultry and humans in Phylogenetic analysis of multidrug-resistant Escherichia coli clones isolated from humans and poultry..

This framework extends down to the cellular level within a single patient. As a tumor grows, individual cells accumulate distinct somatic mutations, effectively evolving like asexual organisms. In Single-cell sequencing data reveal widespread recurrence and loss of mutational hits in the life histories of tumors, Jack Kuipers et al. demonstrated that when single-cell sequencing data allows for the recurrence or loss of mutations, the tree search space expands dramatically. To navigate this complexity, the authors employed MCMC algorithms to sample the most probable non-linear tree structures, proving that probabilistic sampling is as vital for understanding cancer progression as it is for mapping the tree of life.

History & citationVersion 1 · 2026-10-04

Current · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_phylogenetic_tree_construction_maximum_l_98a76d98,
  title = {phylogenetic tree construction (maximum likelihood inference)},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/phylogenetic-tree-construction-maximum-likelihood-inference/rdr_98a76d98dc534c34831468cc7b14dadc},
  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_phylogenetic_tree_construction_maximum_l_98a76d98_v1,
  title = {phylogenetic tree construction (maximum likelihood inference)},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/phylogenetic-tree-construction-maximum-likelihood-inference/rdr_98a76d98dc534c34831468cc7b14dadc/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