Cluster based prediction of PDZ-peptide interactions.

Area of Science:

• Molecular Biology
• Bioinformatics
• Computational Biology

Background:

• PDZ domains are crucial protein recognition modules involved in cellular signaling.
• Existing computational methods for predicting PDZ-peptide interactions often lack broad domain coverage and specificity.
• High-throughput techniques like protein microarrays and phage display provide in vitro binding data but are limited in scope.

Purpose of the Study:

• To develop improved computational models for predicting PDZ domain-peptide interactions with enhanced specificity and broader domain coverage.
• To create specialized binding models for PDZ domain families to improve prediction accuracy for new domains.
• To address the challenge of limited negative data in experimental interaction studies.

Main Methods:

• Clustering of PDZ domains from human, mouse, fly, and worm proteomes to identify 138 novel PDZ families.
• Development of support vector machine (SVM) models for 43 PDZ families with available interaction data.
• Implementation of a semi-supervised technique to generate high-confidence negative interaction data, addressing class imbalance.

Main Results:

• Successfully identified 138 PDZ families comprising 548 PDZ domains.
• Built specialized SVM models for 43 PDZ families, covering 226 PDZ domains.
• Achieved competitive predictive performance compared to state-of-the-art methods, with enhanced domain coverage and specificity.
• Generated genome-wide predictions for human and mouse PDZ domains, uncovering novel biologically relevant interactions.

Conclusions:

• Clustering techniques effectively increase domain coverage for predictive modeling.
• A semi-supervised strategy provides a robust method for generating high-confidence negative data.
• The developed approach allows for high-order correlations and is generalizable to other peptide recognition modules like SH2 domains.
• The freely available predictive models and genome-wide predictions facilitate further research in PDZ domain interactions.