Prediction of protein secondary structure by combining nearest-neighbor algorithms and multiple sequence alignments

Area of Science:

• Computational Biology
• Structural Bioinformatics
• Protein Structure Prediction

Background:

• Predicting protein secondary structure is crucial for understanding protein function.
• Previous methods like neural networks and nearest-neighbor approaches have limitations in accuracy.
• Existing scoring systems do not fully account for specific structural elements or sequence context.

Purpose of the Study:

• To improve the accuracy of protein secondary structure prediction.
• To develop a more refined scoring system that considers N/C-terminal positions and specific secondary structure types.
• To reduce computational time by optimizing the protein database used for prediction.

Main Methods:

• Enhanced a scoring system combining sequence similarity and local structural environment.
• Incorporated N- and C-terminal positions of alpha-helices and beta-strands, and beta-turns.
• Restricted the protein database to a smaller subset of proteins similar to the query sequence.
• Utilized multiple sequence alignments and a jury decision procedure.

Main Results:

• Achieved a sustained overall three-state accuracy of 72.2% for protein secondary structure prediction.
• Demonstrated superior performance compared to the most accurate multilayered neural-network approach on the same dataset.
• Reduced computation time through database restriction.

Conclusions:

• The refined scoring system significantly improves protein secondary structure prediction accuracy.
• Considering specific structural features and optimizing the search database are effective strategies for enhancing prediction.
• The developed method offers a more accurate and efficient alternative for secondary structure prediction.