Partitioned likelihood support and the evaluation of data set conflict

Area of Science:

• Phylogenetics and Evolutionary Biology
• Computational Biology
• Bioinformatics

Background:

• Assessing clade support in phylogenetic analyses of multiple data partitions is crucial for understanding evolutionary relationships.
• Traditional methods may not accurately reflect the contribution of individual data partitions to overall tree support.
• Partitioned Branch Support (PBS) was developed for parsimony-based analyses to address this limitation.

Purpose of the Study:

• To introduce and describe Partitioned Likelihood Support (PLS) as a likelihood-based analogue to PBS.
• To enable the quantification of each data set's contribution to clade support in simultaneous phylogenetic analyses.
• To provide a framework for evaluating phylogenetic signal from individual data partitions within a combined analysis.

Main Methods:

• Simultaneous analysis of all data partitions to obtain the maximum likelihood (ML) tree.
• Identification of the best alternative tree lacking the clade of interest.
• Fitting each data set to both trees and calculating the log-likelihood difference to derive PLS.
• Utilizing resampling methods (e.g., Kishino-Hasegawa test) to assess the statistical significance of PLS.

Main Results:

• PLS quantifies the support for a clade derived from each individual data partition within a combined ML analysis.
• The contribution of each data set to clade support can differ significantly from support observed in separate analyses.
• PLS calculations are feasible irrespective of the complexity of the employed ML model.

Conclusions:

• PLS offers a robust method for evaluating the influence of individual data partitions on phylogenetic inference in likelihood-based analyses.
• This approach enhances the understanding of data set congruence and conflict in phylogenetics.
• Further investigation into the assumptions and appropriateness of significance testing for PLS is warranted.