FUBAR: a fast, unconstrained bayesian approximation for inferring selection.

Area of Science:

• Evolutionary biology
• Computational biology
• Bioinformatics

Background:

• Model-based analyses of natural selection often use a limited number of site classes, imposing constraints that can lead to inaccurate inferences.
• Existing methods may yield misleading results when sites within the same class experience varied selection pressures.

Purpose of the Study:

• To develop a robust and efficient method for analyzing natural selection at the molecular level.
• To overcome model misspecification issues inherent in methods using a small number of site classes.
• To enable faster and more accurate identification of sites under positive and purifying selection.

Main Methods:

• An approximate hierarchical Bayesian method employing a Markov chain Monte Carlo (MCMC) routine.
• Averaging over a large number of predefined site classes to ensure robustness against model misspecification.
• Development of the Fast Unconstrained Bayesian AppRoximation (FUBAR) algorithm.

Main Results:

• FUBAR effectively unconstrains the distribution of selection parameters, providing more accurate analyses.
• The method identifies sites experiencing positive and purifying selection significantly faster than existing approaches.
• FUBAR demonstrates higher statistical power compared to unconstrained fixed-effects likelihood methods.
• Demonstrated applicability on a large influenza hemagglutinin dataset (3,142 sequences).

Conclusions:

• FUBAR offers a robust and computationally efficient alternative for analyzing natural selection.
• The method addresses limitations of traditional site class models, improving inference accuracy.
• FUBAR facilitates the analysis of larger datasets, advancing evolutionary studies.