modPDZpep: a web resource for structure based analysis of human PDZ-mediated interaction networks

Area of Science:

• Bioinformatics
• Structural Biology
• Computational Biology

Background:

• PDZ domains are crucial for biological processes, recognizing short C-terminal sequences in interaction partners.
• Existing bioinformatics tools for PDZ interaction networks rely on machine learning trained on specific PDZ families, limiting their broad applicability.
• Current methods do not facilitate analysis of PDZ-peptide interaction interfaces.

Purpose of the Study:

• To develop a structure-based program, modPDZpep, for predicting human PDZ domain interaction partners.
• To enable the analysis of structural details within PDZ interaction interfaces.
• To provide a tool for genome-scale analysis of PDZ interaction networks.

Main Methods:

• Utilized a structure-based approach to model PDZ-peptide complexes.
• Employed residue-based statistical pair potentials to evaluate binding energy scores.
• Developed modPDZpep, a program that does not require experimental peptide binding affinity data for training.

Main Results:

• modPDZpep predicts interaction partners for diverse PDZ families by evaluating binding energy scores.
• The program is efficient for genome-scale analysis due to a simple scoring function.
• Benchmarking demonstrated good predictive power (ROC-AUC 0.7–0.9) across numerous human PDZ domains.
• modPDZpep can map novel PDZ-mediated interactions using experimental or structure-based predictions.

Conclusions:

• modPDZpep is a novel web-server for structure-based analysis of human PDZ domains.
• The tool overcomes limitations of existing machine learning-based predictors.
• modPDZpep is freely available for research use.