A Bayesian Supertree Model for Genome-Wide Species Tree Reconstruction

Area of Science:

• Phylogenomics
• Computational Biology
• Evolutionary Biology

Background:

• Phylogenomic datasets necessitate species tree methods capable of resolving incongruence from multiple evolutionary processes.
• Existing methods often address single discordance sources like gene duplication/loss or incomplete lineage sorting (ILS).
• There's a need to utilize all available data without requiring ortholog identification or single individuals per species.

Purpose of the Study:

• To develop a species tree inference method that accommodates multiple sources of gene tree/species tree incongruence.
• To create a model that maximizes data utilization from multilocus, genome-wide datasets.
• To implement a robust and efficient computational tool for phylogenetic analysis.

Main Methods:

• Extended Maximum Likelihood (ML) supertrees into a hierarchical Bayesian model.
• Developed the 'guenomu' program to process posterior distributions of gene tree topologies.
• Accounted for gene duplication, loss, and ILS in a modular fashion without ortholog identification.

Main Results:

• The proposed Bayesian supertree method outperformed other approaches in simulated datasets, even without branch lengths.
• The method demonstrated strong performance on empirical data and proved computationally efficient for large datasets.
• The Bayesian approach improved gene tree estimations by reducing topological uncertainty.

Conclusions:

• The novel Bayesian supertree method provides a powerful tool for species tree inference from complex phylogenomic data.
• The 'guenomu' program offers a robust, efficient, and accurate solution for analyzing multilocus sequence data.
• Gene tree parsimony emerges as a competitive alternative under complex scenarios, particularly regarding speed.