Betaturns

Betaturns predicts specific beta-turn types (Types I, II, IV, and VIII) in protein sequences to characterize beta-turn configurations relevant to protein secondary and tertiary structure.

Key Features:

• Type-Specific Prediction: Predicts four distinct beta-turn types: I, II, IV, and VIII.
• Neural Network Architecture: Uses two consecutive feed-forward back-propagation networks with a single hidden layer for each turn type.
• Input Features: The first network is trained on individual sequences and position-specific scoring matrices (PSSM) generated by PSI-BLAST, and the second network receives the first network's output combined with secondary structure information from PSIPRED.
• Training and Validation: Trained and validated on a non-homologous dataset of 426 protein chains using 7-fold cross-validation.
• Performance Metrics: Reports overall accuracies of 74.5%, 93.5%, 67.9%, and 96.5% for Types I, II, IV, and VIII, respectively, with Matthews correlation coefficients (MCC) of 0.29, 0.29, 0.23, and 0.02.

Scientific Applications:

• Tertiary Structure Prediction: Provides beta-turn type information to aid modeling of protein tertiary structures.
• Protein Engineering: Informs modifications to enhance or alter protein function by revealing specific beta-turn configurations.
• Drug Design: Guides development of therapeutic agents by informing folding and stability considerations tied to specific beta-turn conformations.

Methodology:

Two consecutive feed-forward back-propagation neural networks (single hidden layer each) perform sequence-to-structure mapping using individual sequences and PSSMs from PSI-BLAST, with the first network's output combined with PSIPRED secondary-structure predictions as input to a second-level structure-to-structure network; training used a non-homologous set of 426 chains with 7-fold cross-validation.