Analysis and prediction of protein local structure based on structure alphabets

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Machine Learning in Biology

Background:

• Protein structure prediction is crucial for understanding biological function.
• Local structure information has shown promise in advancing prediction accuracy.
• Existing methods often utilize fragment libraries or structure alphabets.

Purpose of the Study:

• To investigate the entropies and correlations of local protein structures.
• To develop and evaluate a novel dual-layer model for protein local structure prediction.
• To compare prediction performance using different representations of structure alphabets.

Main Methods:

• Calculation of entropies and correlations for local protein structures.
• Development of a dual-layer neural network classifier.
• Input features include Position-Specific Iterated BLAST (PSI-BLAST) generated matrices.
• Exploration of two structure alphabets: Cartesian coordinate space and torsion angle space.

Main Results:

• Neighboring local structures exhibit strong correlations.
• The dual-layer model achieved Q-scores of 0.456 and 0.585 for the two structure alphabets.
• These scores represent a significant improvement over existing related methods.

Conclusions:

• The dual-layer model effectively utilizes correlations in local structures for improved prediction.
• The proposed method offers an efficient approach to protein local structure prediction.
• The choice of structure alphabet representation impacts prediction performance.