Context similarity scoring improves protein sequence alignments in the midnight zone.

Area of Science:

• Bioinformatics
• Computational Biology
• Structural Bioinformatics

Background:

• High-quality protein sequence alignments are critical for applications like template-based protein structure prediction.
• Current profile-profile alignment tools often incorporate 1D structural properties (e.g., secondary structure, solvent accessibility) predicted from local sequence windows.
• These additional terms evaluate the conservation of local amino acid properties, capturing correlations between profile columns.

Purpose of the Study:

• To develop a novel, agnostic approach for learning conserved patterns in protein sequences.
• To improve the accuracy of protein sequence alignments by incorporating a new context similarity score.
• To enhance downstream applications reliant on precise pairwise alignments.

Main Methods:

• An unsupervised learning approach was employed to identify maximally conserved patterns.
• A set of 32 conserved patterns, represented by 13-residue sequence profiles, were trained using a maximum likelihood approach.
• These patterns were learned from structurally aligned training profiles of remotely homologous protein pairs.
• The developed context score was integrated into the Hmm-Hmm alignment tool (hhsearch).

Main Results:

• The study successfully learned a set of 32 conserved sequence profiles in an unsupervised manner.
• The integration of the new context score into hhsearch led to significant improvements in the quality of difficult protein sequence alignments.
• The agnostic approach avoids the need to pre-define or predict specific 1D structural properties.

Conclusions:

• The context similarity score demonstrably enhances the quality of protein homology models.
• This improvement extends to other methods that rely on accurate pairwise sequence alignments.
• The findings highlight the utility of learning conserved sequence patterns directly from data for improved alignment accuracy.