Quantitative phylogenomics

This session features two speakers: Hector Banos, Assistant Professor of Mathematics at California State University, whose research focuses on phylogenetic inference and network models, and Benjamin Teo, a Postdoctoral Researcher at the University of Melbourne, working on probabilistic and computational methods for continuous trait evolution on phylogenetic networks. See below for details.

【Talk 1】
Speaker: Hector Banos
Title: Bringing a Knife to a Gunfight: Pitfalls of Phylogenetic Inference under Model Misspecification
Abstract:
Phylogenetic networks provide a flexible framework for representing evolutionary histories that include hybridization, introgression, and other reticulate processes. However, inferring such networks remains computationally and statistically difficult. Many current methods often scale only to restricted classes of networks. Consequently, researchers frequently analyze their data using simpler models (most commonly phylogenetic trees) even when there is strong evidence that the underlying evolutionary history is more complex.

In this talk, we examine the impact of model misspecification on phylogenetic inference, focusing on situations in which data are generated by a complex network but are analyzed using simpler tree or network models. I then show how this mismatch can influence the topology of inferred trees, as well as the structure of inferred networks. These results highlight the limitations and the practical consequences of using simplified models for phylogenetic inference.

【Talk2】
Speaker: Benjamin Teo
Title: Adapting cluster graphs for inference of continuous trait evolution on phylogenetic networks
Abstract:
I consider a new approach ("loopy belief propagation") for fitting Gaussian models on a phylogenetic network to explain the data observed across present-day species for a continuous univariate or multivariate trait. We previously showed [1] that a trait evolution model coupled to a network can be readily cast as a probabilistic graphical model, so that the likelihood can be efficiently computed using a dynamic programming framework ("belief propagation") defined on an auxiliary graph ("cluster graph") that is tree-structured. Even so, maximum likelihood estimation can grow computationally prohibitive for large complex networks.

Belief propagation can be applied more generally to non-tree ("loopy") cluster graphs to compute a factored energy approximation to the log-likelihood. "Loopy" belief propagation may provide a more practical trade-off between estimation accuracy and runtime. However, the influence of cluster graph structure on this trade-off is not precisely understood.

We conduct a simulation study using our Julia package PhyloGaussianBeliefProp [2] to investigate how varying the maximum cluster size of a cluster graph affects this trade-off. We discuss recommended choices for maximum cluster size, and prove the equivalence of likelihood-based and factored-energy based estimates for the homogeneous Brownian motion trait model.

The talk is based on our preprint [3]. I will introduce the key concepts from the ground up.