Accelerated Estimation of Frequency Classes in Site-Heterogeneous Profile Mixture Models

Area of Science:

• Bioinformatics
• Computational Biology
• Evolutionary Biology

Background:

• Proteins exhibit site-specific amino acid preferences due to structural and functional constraints.
• Ignoring this site heterogeneity can introduce artifacts into phylogenetic estimations.
• Existing site-heterogeneous mixture-models often rely on external databases for fixed frequency vectors, limiting their applicability.

Purpose of the Study:

• To develop a novel composite likelihood approach for estimating component frequencies within mixture models.
• To address the limitations of using fixed, externally derived frequency vectors in phylogenetic analyses.
• To improve the accuracy and efficiency of phylogenetic inference using site-heterogeneous models.

Main Methods:

• Proposed a composite likelihood method to estimate mixture model component frequencies directly from the alignment data.
• Investigated necessary adjustments to the composite likelihood for analyses with a limited number of taxa.
• Compared the performance of the proposed method against hierarchical clustering and standard mixture models with fixed components.

Main Results:

• The composite likelihood approach demonstrated significant improvements over hierarchical clustering in simulations.
• Substantial likelihood improvements were observed when applying the method to empirical data compared to fixed-component mixtures.
• The proposed method effectively utilizes alignment-specific data for frequency estimation.

Conclusions:

• The developed composite likelihood approach offers a more accurate and data-driven method for estimating site-heterogeneous mixture model parameters.
• This approach mitigates artifacts in phylogenetic estimation caused by amino acid frequency heterogeneity.
• The method provides a computationally feasible and effective alternative for phylogenetic inference, especially for datasets with fewer taxa.