BhairPred

BhairPred predicts beta-hairpin motifs within protein sequences to identify supersecondary structural elements involved in protein folding and stability.

Key Features:

• SVM-based classifier: Uses a support vector machine for final classification, achieving up to 79.2% accuracy on the reported test set.
• Training dataset: Trained and tested on 5,102 beta-hairpins and 5,131 non-hairpin sequences derived from a non-redundant set of 2,880 proteins.
• Data curation: Dataset annotations were generated using DSSP and PROMOTIF to define secondary-structure-derived hairpins.
• Input features: Incorporates amino acid sequence, PSI-BLAST-derived evolutionary profiles, observed or predicted secondary structure, and surface accessibility as inputs.
• ANN baseline: An artificial neural network baseline produced 65.5% accuracy with sequence alone and 69.1% accuracy after adding PSI-BLAST profiles, secondary structure, and surface accessibility.
• SVM with predicted inputs: The SVM approach reached 77.9% accuracy when predicted secondary structures and surface accessibility were used instead of observed ones.
• External benchmark: Achieved 71.1% accuracy and a Matthew's correlation coefficient (MCC) of 0.41 on a dataset previously analyzed by Cruz et al.
• CASP6 assessment: Performance was evaluated in the context of the CASP6 community-wide experiment on protein structure prediction.

Scientific Applications:

• Protein folding analysis: Identification of beta-hairpins to study folding pathways and supersecondary structure formation.
• Stability assessment: Use of predicted hairpins to analyze contributions to protein stability.
• Functional motif identification: Detection of beta-hairpin motifs that may correspond to functional or interaction-related elements.
• Computational structural annotation: Annotation of structural features for computational biology studies of protein interactions and activity.

Methodology:

Computational methods explicitly stated include an initial artificial neural network and a support vector machine classifier trained/tested on datasets curated with DSSP and PROMOTIF, using PSI-BLAST evolutionary profiles and observed or predicted secondary structure and surface accessibility; evaluations included a Cruz et al. dataset and CASP6.