Less is more: an adaptive branch-site random effects model for efficient detection of episodic diversifying selection

Area of Science:

• Evolutionary biology
• Molecular evolution
• Phylogenetics

Background:

• Comparative sequence analysis uses codon-substitution models to detect natural selection.
• Branch-site models allow selective pressures (ω ratio) to vary across codon sites and phylogenetic branches.

Purpose of the Study:

• To develop and present the adaptive branch-site random effects likelihood (aBSREL) method.
• To improve the efficiency and accuracy of detecting natural selection signatures.

Main Methods:

• Developed aBSREL, a branch-site model with variable parametric complexity selected via an information theoretic criterion.
• Applied aBSREL to simulated data and 8,893 Euteleostomes gene alignments.
• Utilized model selection to identify key branches requiring complex evolutionary modeling.

Main Results:

• aBSREL matches or exceeds existing methods' performance while being an order of magnitude faster.
• Over 80% of branches in typical gene phylogenies can be modeled with a single ω ratio.
• Identified a limited number of key branches where complex evolutionary modeling is essential.

Conclusions:

• aBSREL offers a more efficient and accurate approach to detecting natural selection.
• Many current branch-site models are unnecessarily complex.
• Data-driven model selection is crucial for accurately modeling evolutionary complexity in specific branches.