The promise of composite likelihood for species-level phylogenomic inference

Laura S. Kubatko, Sungsik Kong, Emerson Webb, Zixuan Chen. 2025. Evolutionary Journal of the Linnean Society, kzaf008, DOI:10.1093/evolinnean/kzaf008

Abstract

Species-level phylogenetic inference under the multispecies coalescent model remains challenging in the typical infe- rence frameworks (e.g., the likelihood and Bayesian frameworks) due to the dimensionality of the space of both gene trees and species trees. Summary methods – methods that first estimate gene trees and then use these estimated trees as input data for species tree inference – are limited by the potential effects of gene tree estimation error and by their inapplicability to single nucleotide polymorphism (SNP) data. We extend the earlier work of Peng et al. (2022) and Kong et al. (2024) to develop a composite likelihood framework for species tree estimation. The method can accommodate both SNP and multilocus data, as well as rate variation across sites. Simulation is used to demonstrate that estimates of speciation times on fixed species trees have good statistical properties under these models, such as consistency and asymptotic normality. Simulation is also used to show that species trees can be accurately and efficiently estimated. Throughout, we highlight the potential of composite likelihood to provide efficient estimation of species-level phylogenies in a firm statistical framework.