Phylogenomic mixture models outperform homogeneous and partitioned models.

Area of Science:

• Phylogenetics and evolutionary biology
• Computational biology and bioinformatics

Background:

• Resolving the tree of life faces challenges due to debated phylogenetic nodes.
• Mixture models have emerged to address across-site compositional heterogeneity, improving phylogenetic analyses.
• Skepticism exists regarding the utility of mixture models in phylogenetics.

Purpose of the Study:

• To compare the performance of mixture models against homogeneous and partitioned models.
• To evaluate the accuracy and fit of mixture models for phylogenetic datasets.
• To investigate the effectiveness of CAT-GTR, an infinite mixture model, in phylogenomic analyses.

Main Methods:

• A large-scale simulation study was conducted.
• Mixture models accounting for compositional heterogeneity were compared with homogeneous and partitioned models.
• Model fit and phylogenetic accuracy were assessed.

Main Results:

• Mixture models demonstrated superior fit and accuracy for compositionally heterogeneous datasets.
• The CAT-GTR model, an infinite mixture model, maximized both accuracy and fit.
• Mixture models, including CAT-GTR, performed robustly on homogeneous datasets by converging to appropriate homogeneous models.

Conclusions:

• The study validates the utility of mixture models for handling compositional heterogeneity in phylogenetic analyses.
• CAT-GTR is identified as a highly flexible and accurate model for phylogenomics.
• Results address and dissipate doubts concerning the application of mixture models in the field.