Ancestral sequence reconstruction: accounting for structural information by averaging over replacement matrices.

Area of Science:

• Evolutionary biology
• Structural biology
• Bioinformatics

Background:

• Ancestral sequence reconstruction (ASR) is crucial for understanding protein evolution, structure, and function.
• Existing ASR methods often overlook position-specific evolutionary constraints imposed by protein 3D structure.

Purpose of the Study:

• To develop a novel ASR algorithm that incorporates protein structural information.
• To improve the accuracy of ASR by allowing different protein sites to evolve under distinct evolutionary constraints.

Main Methods:

• Developed a new ASR algorithm allowing variable mixtures of replacement matrices per protein site.
• Assigned replacement matrices based on solvent accessibility, derived from either 3D structures or predicted from sequences.
• Compared the performance of the structure-aware model against standard ASR models using log-likelihoods.

Main Results:

• The structure-aware ASR model achieved higher log-likelihoods than models using a single replacement matrix for all sites.
• Predicting solvent accessibility from sequences yielded improved ASR log-likelihoods compared to using known 3D structures.
• The use of structure-aware mixture models led to substantial differences in the inferred ancestral sequences.

Conclusions:

• Incorporating protein structural information, specifically solvent accessibility, significantly enhances ASR accuracy.
• Predicting structural features from sequences offers a viable alternative for structure-aware ASR when 3D structures are unavailable.
• This novel approach provides more accurate ancestral sequences, advancing the study of protein evolution.