GPS-PBS: A Deep Learning Framework to Predict Phosphorylation Sites that Specifically Interact with...

Area of Science:

• Molecular Biology
• Cellular Signaling
• Bioinformatics

Background:

• Protein phosphorylation regulates cellular activities via specific residue modifications.
• Phosphorylation sites (p-sites) interact with phosphoprotein-binding domains (PPBDs) to propagate signaling.
• Knowledge of PPBD-interacting p-sites is limited, hindering pathway analysis.

Purpose of the Study:

• To develop a computational tool for predicting PPBD-specific binding p-sites (PBSs).
• To identify novel interactions between p-sites and PPBDs in mammalian cells.
• To enhance the understanding of phosphorylation-dependent signaling networks.

Main Methods:

• Collected 4458 known PBSs.
• Improved the group-based prediction system (GPS) algorithm.
• Employed deep learning and transfer learning for model training.
• Developed the GPS-PBS online service for hierarchical prediction.

Main Results:

• GPS-PBS accurately predicts PBSs for 122 PPBD clusters.
• Achieved competitive accuracy compared to existing tools.
• Predicted 371,018 mammalian p-sites potentially interacting with PPBDs.
• Revealed co-regulation of p-sites by PPBD-containing proteins and protein kinases.

Conclusions:

• GPS-PBS is a valuable tool for predicting PPBD-p-site interactions.
• Facilitates the dissection of complex phosphorylation signaling networks.
• Highlights the coordinated regulation of signaling pathways by multiple proteins.