PDZPepInt

PDZPepInt predicts binding peptides of PDZ domains across humans, mice, flies, and worms using clustering and a semi-supervised strategy to improve domain coverage and prediction confidence.

Key Features:

• Cluster-Based Prediction: Utilizes clustering techniques to increase PDZ domain coverage for comprehensive prediction.
• Peptide Representation: Represents peptides as 5 C-terminal residues of binding proteins for prediction and analysis.
• Intrinsically Unstructured Peptides: Considers C-terminal peptides that are intrinsically unstructured when identifying interactions.
• Gene Ontology Integration: Integrates the Gene Ontology database to retrieve reliable interaction evidence.
• High-Order Correlations: Incorporates high-order correlations between amino acid positions in binding peptides to enhance accuracy.
• Semi-Supervised Negative Data: Employs a semi-supervised strategy to obtain high-confidence negative interaction data.
• General Applicability: Applies the same approach to other peptide recognition modules, for example SH2 domains.
• Genome-Wide Predictions: Performs genome-wide predictions, including for 101 human and 102 mouse PDZ domains.

Scientific Applications:

• Domain Coverage Expansion: Increasing PDZ domain coverage through clustering to enable broader interaction mapping.
• Genome-Wide Interaction Mapping: Predicting PDZ–peptide interactions across human and mouse proteomes at genome scale.
• Negative Dataset Construction: Generating high-confidence negative interaction datasets via semi-supervised methods.
• Correlation Analysis: Analyzing high-order amino acid position correlations to elucidate determinants of PDZ binding specificity.
• Cross-Module Extension: Extending prediction methodology to other peptide recognition modules such as SH2 domains.

Methodology:

Uses clustering techniques; employs a semi-supervised strategy to obtain high-confidence negatives; represents peptides as 5 C-terminal residues and considers intrinsically unstructured C-terminal peptides; integrates Gene Ontology for interaction retrieval; models high-order correlations between amino acid positions; performs genome-wide predictions for 101 human and 102 mouse PDZ domains.