Modeling Site Heterogeneity with Posterior Mean Site Frequency Profiles Accelerates Accurate Phylogenomic Estimation

Area of Science:

• Computational Biology
• Phylogenetics
• Molecular Evolution

Background:

• Protein evolution exhibits site-specific amino acid preferences due to structural and functional constraints.
• Common models fail to capture this heterogeneity, potentially causing phylogenetic errors like long-branch attraction.
• Existing site-heterogeneous mixture models are computationally intensive, limiting their application in large phylogenomic studies.

Purpose of the Study:

• To introduce a rapid and efficient method, Posterior Mean Site Frequency (PMSF), as an approximation for empirical profile mixture models in maximum likelihood (ML) phylogenetics.
• To enable accurate phylogenetic inference on large datasets by overcoming the computational limitations of existing models.

Main Methods:

• Developed the PMSF method, which assigns a conditional mean amino acid frequency profile to each site.
• Integrated PMSF into IQ-TREE for efficient phylogenetic tree searching.
• Compared PMSF performance against traditional mixture models using simulations and empirical data.

Main Results:

• The PMSF method significantly accelerates computation (approx. k/1.5-fold) and reduces memory requirements compared to mixture models.
• PMSF effectively mitigates long-branch attraction artefacts in phylogenetic estimations.
• Enabled, for the first time, full nonparametric bootstrap analyses under complex site-heterogeneous models for large datasets.

Conclusions:

• The PMSF method offers a computationally feasible and accurate alternative for modeling site-heterogeneous protein evolution.
• It enhances the reliability of phylogenetic reconstructions, particularly for large-scale phylogenomic analyses.
• PMSF provides more accurate phylogenetic estimates than the full mixture models in certain scenarios.