Application of protein structure alignments to iterated hidden Markov model protocols for structure prediction

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Bioinformatics

Background:

• Protein structure prediction from sequence is crucial for understanding protein function.
• Profile hidden Markov models (HMMs) are powerful tools, built using sequence alignments.
• Accurate alignment of diverse or distantly related sequences remains a challenge for sequence-only methods.

Purpose of the Study:

• To investigate the utility of incorporating protein structure alignments into iterative profile hidden Markov model generation.
• To benchmark structure alignment-enhanced models against sequence-only models for protein structure prediction.

Main Methods:

• Exploration of iterative protocols for profile hidden Markov model generation.
• Inclusion of protein structure alignments within these protocols.
• Large-scale creation and benchmarking of structure alignment-enhanced models.

Main Results:

• Structure alignment-enhanced models did not universally outperform sequence-only models at the superfamily level.
• Structure alignment-enhanced models complemented sequence-only models, particularly in the twilight zone of sequence identity.
• Combining both model types improved structure prediction accuracy over sequence-only models alone.
• Structure alignment-enhanced models yielded superior fold-level structure assignments.

Conclusions:

• A combined approach using both traditional sequence-only models and structure alignment-enhanced models is recommended for predicting the structure of remote homologs.
• This hybrid strategy offers enhanced performance, especially for challenging predictions.