A neural network method for prediction of beta-turn types in proteins using evolutionary information

Area of Science:

• * Bioinformatics
• * Computational Biology
• * Structural Biology

Background:

• * Protein secondary structure prediction is crucial for understanding protein function.
• * Existing methods like Betatpred2 predict general beta-turns but lack type-specific classification.
• * Predicting specific beta-turn types (I, II, IV, VIII) is essential for accurate tertiary structure prediction.

Purpose of the Study:

• * To develop a computational method for predicting specific beta-turn types (I, II, IV, VIII).
• * To improve upon existing beta-turn prediction methods by incorporating type-specific information.
• * To enhance the accuracy of protein tertiary structure prediction through precise beta-turn type identification.

Main Methods:

• * Development of a prediction method utilizing feed-forward back-propagation networks.
• * Training networks on single sequences and Position-Specific Scoring Matrices (PSSM) from PSI-BLAST.
• * Integration of predicted secondary structure information from PSIPRED and a second-level network for enhanced accuracy.

Main Results:

• * Achieved significant prediction performance improvements for beta-turn types I, II, IV, and VIII using multiple sequence alignment.
• * Demonstrated further performance gains with a second-level network and PSIPRED secondary structure information.
• * Reported prediction accuracies of 74.5% (Type I), 93.5% (Type II), 67.9% (Type IV), and 96.5% (Type VIII) with corresponding MCC values.

Conclusions:

• * The developed method accurately predicts specific beta-turn types, outperforming random prediction.
• * Type I and II beta-turns exhibit better prediction performance compared to Type IV and VIII.
• * The approach provides valuable insights for tertiary structure prediction and is accessible via a web server.