Generalised Bayesian distance-based phylogenetics for the genomics era

Matthew J. Penn, Neil Scheidwasser, Mark P. Khurana, Christl A. Donnelly, David A. Duchêne, Samir BhattarXiv 2025 · DOI · arXiv

TL;DR

The paper introduces a novel "entropic likelihood" method for phylogenetic inference that bridges the gap between computationally intensive likelihood-based methods and faster but less accurate distance-based methods. This new approach enables efficient Bayesian inference on massive genomic datasets, a critical need in the genomics era.

What Problem Does This Paper Address?

Phylogenetics, the study of evolutionary relationships, traditionally relies on methods like Maximum Likelihood (ML) and Bayesian inference. However, as genomic datasets grow exponentially in size (millions of sites, thousands of taxa), these methods become computationally prohibitive. Distance-based methods, while fast, have historically been criticized for lower accuracy due to their inability to fully account for evolutionary processes across the entire tree. This paper addresses the challenge of performing accurate and efficient phylogenetic inference on the massive datasets characteristic of the "genomics era."

Methods

The core innovation is the development of an "entropic likelihood." This method leverages the concept of tree entropy, which quantifies the uncertainty in evolutionary processes, to create a likelihood function that is computationally tractable for large datasets. The authors demonstrate a strong theoretical link between this entropic likelihood and Felsenstein's classic likelihood, suggesting comparable performance. They achieve this by:

  • Defining an "entropic distance matrix": This matrix captures evolutionary distances based on entropy, derived from genetic distances.
  • Developing an entropic likelihood function: This function is designed to be efficiently computable, particularly for Bayesian inference.
  • Establishing a linear relationship: The entropic likelihood is shown to be linearly related to both the Balanced Minimum Evolution (BME) objective and Felsenstein's likelihood, allowing for calibration and approximate inference.
  • Utilizing Bayesian inference: The entropic likelihood is integrated into a Bayesian framework, enabling the estimation of posterior probabilities for tree topologies and branch lengths.

Datasets

The authors evaluated their method using:

  • Three phylogenetic benchmark datasets:
    • Tetrapods (Hedges et al., 1990): 27 taxa, 1949 sites.
  • Acanthocephalans (Garey et al., 1996): 29 taxa, 2520 sites.
  • Mammals (Yang and Yoder, 2003): 36 taxa, 1812 sites.
  • A large-scale avian dataset: The Bird 10,000 Genomes (B10K) project dataset, comprising 363 genomes (representing 92% of bird families) with a 60-million-site alignment. This dataset highlights the method's scalability for truly genomic-scale phylogenomics.

Key Insights and Ideas

  • Bridging the speed-accuracy gap: The entropic likelihood offers a compelling solution to the long-standing trade-off between computational speed and phylogenetic accuracy.
  • Theoretical link to established methods: Demonstrating a strong correlation with Felsenstein's likelihood and a close relationship with BME provides theoretical justification and practical utility for the entropic approach.
  • Bayesian inference for large datasets: The method unlocks the power of Bayesian inference for datasets previously intractable, allowing for robust uncertainty quantification.
  • Uncertainty in avian diversification: The application to avian genomes revealed significant uncertainty in diversification events immediately following the K-Pg transition, underscoring the method's ability to uncover complex evolutionary histories.

Key Results

  • Computational efficiency: The entropic likelihood method scales orders of magnitude faster than Felsenstein's likelihood, especially with increasing numbers of sites. For the 60-million-site avian dataset, Felsenstein's likelihood would be prohibitively slow.
  • Comparable accuracy: The entropic likelihood, after calibration, yields topological accuracies comparable to maximum likelihood methods, as demonstrated by the avian dataset analysis (91% to 98% topological accuracy).
  • Robust Bayesian inference: The method facilitates Bayesian MCMC sampling, producing posterior distributions that overlap with distance-based bootstraps and are more concentrated than BME optima.
  • Insights into avian evolution: The analysis of avian genomes revealed substantial uncertainty in early avian diversification, particularly around the K-Pg boundary, highlighting the method's power in resolving complex evolutionary events.

Limitations and Future Work

  • Model misspecification: While the entropic likelihood is shown to be robust, the paper acknowledges that approximations can introduce some error, particularly when evolutionary rates are high. Calibration against Felsenstein's likelihood is proposed to mitigate this.
  • Non-ultrametric trees: The current method is primarily designed for ultrametric trees, though the authors suggest it could be extended to non-ultrametric models.
  • Further exploration of branching processes: The authors express interest in examining more complicated branching process models for approximating entropic distances.
  • Incorporating uncertainty in distance matrices: The current analysis uses a single distance matrix; propagating uncertainty from this matrix could lead to larger posterior distributions of trees.

Summary updated

Abstract

As whole genomes become widely available, maximum likelihood and Bayesian phylogenetic methods are demonstrating their limits in meeting the escalating computational demands. Conversely, distance-based phylogenetic methods are efficient, but are rarely favoured due to their inferior performance. Here, we extend distance-based phylogenetics using an entropy-based likelihood of the evolution among pairs of taxa, allowing for fast Bayesian inference in genome-scale datasets. We provide evidence of a close link between the inference criteria used in distance methods and Felsenstein's likelihood, such that the methods are expected to have comparable performance in practice. Using the entropic likelihood, we perform Bayesian inference on three phylogenetic benchmark datasets and find that estimates closely correspond with previous inferences. We also apply this rapid inference approach to a 60-million-site alignment from 363 avian taxa, covering most avian families. The method has outstanding performance and reveals substantial uncertainty in the avian diversification events immediately after the K-Pg transition event. The entropic likelihood allows for efficient Bayesian phylogenetic inference, accommodating the analysis demands of the genomic era.
BibTeX
@misc{penn2025generalised,
  author = {Matthew J. Penn and Neil Scheidwasser and Mark P. Khurana and Christl A. Donnelly and David A. Duchêne and Samir Bhatt},
  title = {{Generalised Bayesian distance-based phylogenetics for the genomics era}},
  year = {2025},
  eprint = {2502.04067},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2502.04067},
  howpublished = {arXiv},
}
Figure