PPIIPRED

PPIIPRED predicts the propensity of polyproline II helix (PPIIH) secondary structure in proteins from primary amino acid sequences to identify PPIIH motifs involved in peptide binding and disordered regions.

Key Features:

• Bidirectional Recurrent Neural Network (Bi-RNN): Employs a bidirectional recurrent neural network trained on protein sequences mapped to known three-dimensional structures to capture sequence dependencies for PPIIH propensity prediction.
• Dihedral Angle Filtering: Incorporates dihedral angle filtering during training to enhance recognition of PPIIH-specific structural characteristics.
• Amino Acid Preferences: Identifies residues that favor PPIIH formation beyond proline, including leucine (Leu), methionine (Met), lysine (Lys), arginine (Arg), glutamic acid (Glu), glutamine (Gln), alanine (Ala), and valine (Val).

Scientific Applications:

• Disordered Protein Regions: Characterizes short PPIIH motifs within intrinsically disordered regions that mediate protein–protein interactions and signaling.
• Evolutionary Analyses: Enables large-scale analysis of PPIIH prevalence and variation across species for evolutionary studies.
• Peptide Binding Studies: Prioritizes candidate binding peptides by predicting PPIIH propensity to reduce datasets for experimental validation and peptide-based discovery.

Methodology:

Training a bidirectional recurrent neural network on protein sequences with known three-dimensional structures with dihedral angle filtering applied during training to refine PPIIH propensity predictions.