PredCSF

PredCSF predicts conotoxin superfamilies from amino acid sequences to classify disulfide-rich peptides relevant to neuronal receptors and ion channels.

Key Features:

• Feature Integration: Incorporates physicochemical properties, evolutionary information, predicted secondary structures, and amino acid composition as input features.
• Classification Algorithm: Implements a modified one-versus-rest Support Vector Machine for multi-class conotoxin prediction.
• Feature Selection: Uses random forest-based feature selection to identify and prioritize the most informative features.
• Performance: Achieves an overall accuracy of 90.65% on a benchmark dataset comprising four main conotoxin superfamilies and one non-conotoxin class.
• Dataset Validation: Validated using the most recent database available for conotoxins.

Scientific Applications:

• Conotoxin characterization: Classifies conotoxin sequences into superfamilies to support studies of their biological and pharmacological functions.
• Peptide discovery and drug development: Facilitates identification of novel conotoxins to accelerate peptide-based drug discovery targeting neuronal receptors and ion channels.

Methodology:

Uses a modified one-versus-rest Support Vector Machine integrating physicochemical properties, evolutionary information, predicted secondary structure, and amino acid composition, with random forest-based feature selection.