Looking for trees in the forest: summary tree from posterior samples

Area of Science:

• Computational Biology
• Phylogenetics
• Evolutionary Biology

Background:

• Bayesian phylogenetic analysis yields multiple trees, necessitating representative summary trees.
• Existing methods for unrooted trees are more developed than those for rooted time trees from BEAST analyses.
• Limited research exists on simultaneously optimizing topology and branch lengths for summary trees.

Purpose of the Study:

• To evaluate and compare existing and novel methods for generating summary trees from Bayesian phylogenetic analyses.
• To assess the performance of summary tree methods for rooted time trees, specifically those from BEAST.
• To provide guidance on selecting appropriate summary tree methods based on user objectives.

Main Methods:

• Empirical comparison of new and established summary tree generation techniques.
• Development of methods based on tree metrics and practical tree concepts.
• Extensive simulations to assess summary tree quality using model fit and accuracy to true trees.

Main Results:

• No single summary tree method excels in all aspects; trade-offs exist between divergence time accuracy and branch length precision.
• New methods utilize more posterior information compared to existing approaches that discard data.
• Performance was evaluated based on explanatory power for sequence data and proximity to simulated true trees.

Conclusions:

• The choice of summary tree method is contingent on the specific research question and desired outcome.
• Users can select the most suitable method by considering the trade-offs highlighted in the study.
• This work aids researchers in making informed decisions for constructing summary trees in phylogenetic studies.