MetaSSPred

MetaSSPred predicts protein secondary structure by integrating 33 physicochemical properties into 15-tuple peptides using Chou's general PseAAC and combining GA-optimized support vector machines with SPINE X to achieve balanced accuracy across helices (H), beta sheets (E), and coils (C).

Key Features:

• PseAAC encoding: Integrates 33 physicochemical properties into 15-tuple peptides using Chou's general PseAAC.
• Binary SVM classifiers: Employs three separate support vector machines for E versus non-E (E/¬E), C/¬C, and H/¬H binary classifications.
• Genetic algorithm optimization: Uses a genetic algorithm to optimize the SVMs and assemble a multiclass predictor (cSVM) that balances precision and recall across classes.
• Integration with SPINE X: Combines cSVM outputs with SPINE X predictions to improve overall secondary structure accuracy, particularly beta sheets (E).
• Performance on benchmark datasets: Achieved QE (beta accuracy) of 71.7% on CB471 and 74.4% on N295, representing improvements of 20.9% and 19.0% over SPINE X, respectively.
• Reduced inter-class variability: Shows lower standard deviations in accuracies across SS classes (4.2% and 2.3% for MetaSSPred versus 12.9% and 10.9% for SPINE X on CB471 and N295, respectively).

Scientific Applications:

• Protein secondary structure prediction: Provides balanced predictions of helices (H), beta sheets (E), and coils (C) for sequence-based structural inference.
• Structural biology and protein folding studies: Supports analyses of tertiary structure and functional interpretation by improving beta-sheet (E) prediction accuracy.

Methodology:

Sequences are encoded as 15-tuple peptides with 33 physicochemical properties via Chou's general PseAAC, classified by three binary SVMs (E/¬E, C/¬C, H/¬H) optimized with a genetic algorithm to form a cSVM multiclass predictor, and integrated with SPINE X outputs.