Exploring the extremes of sequence/structure space with ensemble fold recognition in the program Phyre.

Area of Science:

• Computational Biology
• Structural Biology
• Bioinformatics

Background:

• Genomic sequence annotation is a major challenge in modern biology.
• Protein structure prediction via remote homology detection is a key annotation technique.
• Existing single algorithms struggle with the evolutionary divergence of remote homologues.

Purpose of the Study:

• To develop and evaluate an in-house ensemble prediction method for improved protein structure annotation.
• To investigate the underlying reasons for the superiority of ensemble methods in fold recognition.

Main Methods:

• Implemented a meta-server prediction method (Phyre) using an ensemble of diverse predictive algorithms.
• Applied a concept from protein loop energetics to 3D clustering for improved recognition.
• Developed a stringent test simulating scenarios where single algorithms fail to provide confident assignments.

Main Results:

• Phyre achieved 64.0% accuracy in identifying homologous query-template relationships at 95% precision.
• 84.0% of test proteins were accurately annotated using the Phyre method.
• This represents a 29.6% increase in correct relationships and 46.2% increase in annotated queries compared to single best algorithms over PSI-Blast.

Conclusions:

• Ensemble predictions, like Phyre, offer significant improvements over individual methods in protein structure prediction.
• The power of ensemble methods stems from effective noise reduction and filtering of false positives.
• This approach enhances sequence space coverage, improves model quality, and can reduce experimental workload in structural genomics.