Evaluating Fast Maximum Likelihood-Based Phylogenetic Programs Using Empirical Phylogenomic Data Sets.

Area of Science:

• Phylogenetics and Evolutionary Biology
• Computational Biology
• Genomics

Background:

• Phylogenomic datasets are rapidly growing, necessitating faster and more accurate computational inference methods.
• Several maximum likelihood programs (RAxML/ExaML, PhyML, IQ-TREE, FastTree) are widely used but lack systematic performance comparisons on empirical data.

Purpose of the Study:

• To systematically evaluate and compare the performance of four popular phylogenetic inference programs (RAxML/ExaML, PhyML, IQ-TREE, FastTree).
• To assess program performance based on likelihood maximization, tree topology accuracy, and computational speed using large-scale empirical phylogenomic datasets.

Main Methods:

• Evaluated four programs (RAxML/ExaML, PhyML, IQ-TREE, FastTree) on 19 empirical phylogenomic datasets.
• Datasets varied in size (hundreds to thousands of genes) and taxa number (up to 200).
• Assessed performance metrics including likelihood, tree topology, and computational speed for both single-gene and concatenation-based analyses.

Main Results:

• Exhaustive searches (10 trees) outperformed faster searches (1 tree) for single-gene inference across RAxML, PhyML, and IQ-TREE.
• IQ-TREE and RAxML/ExaML achieved the highest likelihoods in concatenation-based species tree inference.
• FastTree was fastest but yielded lower likelihoods and less consistent topologies; PhyML often failed concatenation analyses.

Conclusions:

• Program performance is influenced by data matrix properties like taxa number and phylogenetic signal strength.
• IQ-TREE and RAxML/ExaML are recommended for accurate and efficient concatenation-based species tree inference.
• Results provide benchmarks to guide large-scale phylogenomic data analysis design.