morePhyML: improving the phylogenetic tree space exploration with PhyML 3.

Area of Science:

• Phylogenetics and evolutionary biology
• Computational biology and bioinformatics

Background:

• Maximum Likelihood (ML) phylogenetic tree inference is crucial for understanding evolutionary relationships.
• PhyML is a popular software for ML phylogenetic tree construction, utilizing hill-climbing methods.
• Tree space exploration is key to finding accurate phylogenetic trees, with Nearest Neighbor Interchange (NNI) and Subtree Pruning and Regrafting (SPR) being common techniques.

Purpose of the Study:

• To evaluate the performance of PhyML's NNI and SPR tree swapping techniques in phylogenetic tree inference.
• To develop and assess a novel method, morePhyML, for improving ML phylogenetic tree accuracy by overcoming local optima.

Main Methods:

• Comparative analysis of PhyML's NNI and SPR tree swapping strategies on various datasets.
• Implementation of a character reweighting strategy, termed ratchet, within a script named morePhyML.
• Benchmarking morePhyML against other recent ML tree inference software using amino acid and nucleotide data.

Main Results:

• PhyML generally infers more likely trees using SPR compared to NNI.
• SPR-based local searches in PhyML frequently fail to identify the optimal ML tree.
• The morePhyML script, employing character reweighting, demonstrates improved tree likelihood estimates.
• morePhyML outperforms several other recent ML tree inference software in accuracy for many datasets.

Conclusions:

• SPR is superior to NNI in PhyML for likelihood but still prone to local optima.
• The morePhyML script offers a significant improvement in phylogenetic tree inference accuracy.
• Character reweighting (ratchet) is an effective strategy for enhancing ML tree search in phylogenetics.