2025/01/01 by Wenjie Zhu, Sebastian Höhna · 1 voice
Biochemistry, Genetics and Molecular Biology · #Genomics and Phylogenetic Studies #Genetic diversity and population structure #Gene expression and cancer classification
paper · pdf · doi:10.1093/evolinnean/kzaf018
openalex publication_date 2025/01/01 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/29
Abstract Gene tree–species tree discordance is widely recognized, and the multispecies coalescent (MSC) model has emerged as the standard framework for species tree inference from multilocus data. However, existing methods face challenges: joint inference methods are computationally intensive or even infeasible for large datasets, while summary methods use approximations and often ignore gene tree estimation uncertainties. Here, we introduce and explore a two-step full-likelihood method combining the efficiency of summary methods with the statistical rigour of full-likelihood approaches. Specifically, our new approach uses either estimated gene trees (RevBayes-MSC) or posterior samples of gene trees (RevBayes-IS) and computes the likelihood under the MSC process without approximations. Through simulation, we evaluated our method against existing approaches under varying levels of incomplete lineage sorting (ILS) and gene tree estimation error. Joint estimation (using BPP) consistently outperformed two-step methods, particularly under high levels of ILS. Among two-step approaches, methods utilizing branch lengths (Maximum Tree and RevBayes-MSC) achieved superior accuracy with known gene trees, while summary methods using only topological information (MP-EST and ASTRAL) demonstrated greater robustness to gene tree uncertainty and estimation errors. Empirical analysis of gibbon data revealed conflicting phylogenetic signals, with our two RevBayes methods and BPP supporting different topologies. Our importance sampling approach (RevBayes-IS) represents a novel full-likelihood methodology for MSC inference that shows promise with increasing sample sizes approaching joint inference performance, though computational efficiency requires further improvement through implementation optimization. Most importantly, this study demonstrates the strength and weakness of using two-step full-likelihood species tree inference.