Bayesian mixture modeling using a hybrid sampler with application to protein subfamily identification

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Predicting protein function is crucial for understanding biological processes.
• Protein families often exhibit functional diversification, making accurate classification challenging.

Purpose of the Study:

• To develop a novel Bayesian mixture approach for discovering functional diversification within protein families.
• To model protein families as mixtures of profile hidden Markov models.

Main Methods:

• Utilized a hybrid Markov chain Monte Carlo sampler with Gibbs sampling and hierarchical clustering for posterior inference.
• Employed integrated likelihoods for mixture size inference.
• Investigated the impact of prior choices on model performance.

Main Results:

• Demonstrated accurate identification of mixture size in simulations, even with homogeneous or mixed data.
• Showcased the method's applicability using two real protein sequence datasets.

Conclusions:

• The proposed Bayesian mixture approach effectively identifies functional diversification in protein families.
• Prior selection is critical for the method's success, with independent data-based priors performing well.