FuncPhos-STR: An integrated deep neural network for functional phosphosite prediction based on AlphaFold protein...

Area of Science:

• Biochemistry
• Bioinformatics
• Structural Biology

Background:

• Phosphorylation is crucial for biological regulation, but assigning functions to most identified phosphosites is challenging.
• Understanding phosphosite function is vital for deciphering cellular mechanisms and disease pathways.

Purpose of the Study:

• To develop FuncPhos-STR, a deep learning framework for predicting human phosphosite function and visualizing their structural context.
• To create an accessible online resource for researchers studying protein phosphorylation.

Main Methods:

• Integrated structural and dynamics data from AlphaFold protein structures with sequence evolution and protein-protein interaction (PPI) information.
• Developed a deep learning model (FuncPhos-STR) building upon the FuncPhos-SEQ framework.
• Evaluated model performance using AUC, achieving a best value of 0.855.

Main Results:

• FuncPhos-STR effectively integrates multi-dimensional features (structural, dynamic, evolutionary, PPI) for enhanced phosphosite function prediction.
• Structural features, particularly phosphosites within pocket regions, correlate with higher functional scores, suggesting regulatory roles.
• The model demonstrates superior performance compared to existing methods.

Conclusions:

• FuncPhos-STR accelerates the functional annotation of uncharacterized phosphosites.
• The framework facilitates the elucidation of allosteric regulation mechanisms mediated by phosphosites.
• The online resource provides a valuable tool for the scientific community to explore phosphosite functions.