Integrating Over Phylogenetic Uncertainty in Evolutionary Models

This direction is merged into another direction: phylogenetic uncertainty. Its last version stays readable and citable here.

Version 1 of 115 papers22 conceptsfrozen 2026-10-04

In most evolutionary studies, the phylogenetic tree—the branching diagram representing the history of species or genes—is treated as a fixed foundation. Researchers typically reconstruct the "best" tree from genetic data and then use that single topology to estimate everything else: how quickly a virus is spreading, when a protein gained a new function, or how many species exist in an ecosystem.

However, this "fixed tree" shortcut is fundamentally risky. Genetic data is often noisy or incomplete, meaning many different tree structures could explain the same data. When we ignore this phylogenetic uncertainty, we don't just get slightly less precise answers; we often get answers that are confidently wrong.

The Problem with the "Best" Tree

The traditional workflow assumes that the error in tree reconstruction is negligible compared to the signal in the downstream analysis. Mathieu Fourment and Marc A. Suchard demonstrated in Assessing the Validity of the Fixed Tree Topology Assumption in Phylodynamic Inference that this assumption is often false. By fixing a single topology, researchers introduce systematic biases into estimates of mutation rates and population growth. The computational speed gained by using a fixed tree comes at the cost of distorting the very demographic history the researcher is trying to uncover.

This issue spans scales, from global pandemics to individual cells. In cell lineage tracing, Jean Feng and Amy D. Willis note in Estimation of cell lineage trees by maximum-likelihood phylogenetics that failing to provide a measure of confidence for specific branches leaves developmental biologists with an incomplete picture of how tissues form.

From Point Estimates to Probability Distributions

The solution is to stop treating the tree as a single point in space and start treating it as a probability distribution. Instead of asking, "What does this tree tell us?" we ask, "What do all plausible trees tell us, weighted by how likely they are?"

This approach, known as integrating over phylogenetic uncertainty, allows errors in one tree to be balanced by the signals in others. Noah M. Reid applied this logic to species delimitation in Phylogenetic estimation error can decrease the accuracy of species delimitation. By averaging results across a posterior distribution of trees, the model provides a more cautious and realistic assessment of biodiversity, preventing the over-counting of species that occurs when a single, potentially incorrect tree is used.

Similarly, Barbara Engelhardt and Michael I. Jordan developed SIFTER (Protein Molecular Function Prediction by Bayesian Phylogenomics), which uses Bayesian inference to predict protein function. By modeling how molecular roles evolve across the tree while accounting for the fact that the tree itself might be wrong, they achieved significantly higher accuracy than methods that rely on simple sequence similarity.

Mechanisms for Handling Uncertainty

How do we practically manage the astronomical number of possible trees? Researchers have developed several mathematical and computational frameworks to make this tractable:

  1. Optimal Transport: Steven N. Evans showed in The phylogenetic Kantorovich-Rubinstein metric for environmental sequence samples that comparing microbial communities can be framed as "Earth Mover's Distance" on a tree. This allows researchers to accommodate probability distributions that represent sequences spread across multiple potential branches rather than a single fixed point.
  2. Probabilistic Programming: To handle complex models where lineages frequently go extinct, Jan Kudlicka and Thomas B. Schön utilized Probabilistic Programming for Birth-Death Models. Their use of an "Alive Particle Filter" with delayed sampling allows for efficient inference without assuming the branching structure is pre-determined.
  3. Compact Data Structures: When thousands of trees are equally plausible, we cannot list them all. Will Dumm introduced the history sDAG in Representing and extending ensembles of parsimonious evolutionary histories with a directed acyclic graph. This structure compactly stores an ensemble of plausible histories, providing a "skeleton" for quantifying uncertainty in massive datasets like SARS-CoV-2.

Why It Works: The Robustness of the Ensemble

The core intuition is that while a single "best" tree might be led astray by a few noisy genetic markers, the ensemble of likely trees will generally converge on the true evolutionary signal.

In Bayesian Hidden Markov Tree Models for Clustering Genes, Yang Li and colleagues demonstrated that treating the phylogenetic structure as a distribution prevents overfitting to a single, potentially incorrect species tree. This makes the clustering of genes far more robust to the "noise" of incomplete genomic data.

By moving away from the "one true tree" paradigm, these methods transform phylogenetic uncertainty from a source of error into a quantifiable part of the biological signal. Whether predicting immune-escape mutations (Daniel Habermann et al., HAMdetector) or inferring population dynamics (Lorenzo Cappello and Julia A. Palacios, adaPop), accounting for the probabilistic nature of history leads to more accurate and honest science.

Papers in this version

History & citationVersion 1 · 2026-10-04

Merged into another direction · Permanent link

View frozen version 1

Cite this direction

Latest version (a living page):

@misc{lacuna_integrating_over_phylogenetic_uncertaint_a1e182c2,
  title = {Integrating Over Phylogenetic Uncertainty in Evolutionary Models},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/integrating-over-phylogenetic-uncertainty-in-evolutionary-models/rdr_a1e182c2463a4bb1a17b6b9876cf2cd6},
  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_integrating_over_phylogenetic_uncertaint_a1e182c2_v1,
  title = {Integrating Over Phylogenetic Uncertainty in Evolutionary Models},
  author = {{Lacuna}},
  howpublished = {Lacuna research direction},
  year = {2026},
  url = {https://lacuna.tiptreesystems.com/direction/integrating-over-phylogenetic-uncertainty-in-evolutionary-models/rdr_a1e182c2463a4bb1a17b6b9876cf2cd6/v/1},
  note = {Version 1, frozen 2026-10-04. Accessed 2026-11-20},
}