Information Criteria for Comparing Partition Schemes.

Area of Science:

• Computational Biology
• Phylogenetics
• Bioinformatics

Background:

• Phylogenetic inference requires decisions on sharing nucleotide substitution parameters across data partitions.
• Partitioning errors, such as lumping heterogeneous data or splitting homogeneous data, can impact phylogenetic accuracy.
• Current methods like the Akaike information criterion (AIC) use penalties based on parameter count, which may not adequately address lumping vs. splitting error asymmetry.

Purpose of the Study:

• To develop adjusted information criteria, AIC^(p) and BIC^(p), that better account for the differential impact of lumping and splitting errors in phylogenetic partitioning.
• To evaluate the performance of these new criteria against standard AIC and BIC in selecting optimal partitioning schemes.

Main Methods:

• Derivation of AIC^(p) and BIC^(p) with adjustments favoring fewer lumping errors.
• Simulation studies to compare the behavior of AIC, BIC, AIC^(p), and BIC^(p) under various partitioning scenarios.
• Empirical data analysis using real biological datasets to assess practical performance.

Main Results:

• AIC^(p) and BIC^(p) were shown to be more robust against lumping errors compared to standard AIC and BIC.
• Simulations and empirical analyses demonstrated the utility of the proposed adjustments in selecting more appropriate partitioning schemes.
• The adjusted criteria showed a reduced probability of making detrimental lumping errors.

Conclusions:

• The proposed AIC^(p) and BIC^(p) offer improved strategies for phylogenetic partitioning by mitigating the risks associated with lumping errors.
• These adjusted criteria provide a more nuanced approach to model selection in phylogenetics, leading to potentially more accurate evolutionary reconstructions.
• The robustness of AIC^(p) and BIC^(p) to model parameterization suggests their broad applicability in phylogenetic analyses.