GraphPhos: Predict Protein-Phosphorylation Sites Based on Graph Neural Networks

Area of Science:

• Biochemistry
• Bioinformatics
• Computational Biology

Background:

• Protein phosphorylation is a crucial post-translational modification regulating numerous cellular processes.
• Accurate identification of phosphorylation sites is essential for understanding protein function and signaling pathways.

Purpose of the Study:

• To develop an advanced computational model for predicting protein phosphorylation sites.
• To improve the accuracy and efficiency of phosphorylation site identification using integrated features.

Main Methods:

• Developed GraphPhos, a graph neural network (GNN) model for phosphorylation site prediction.
• Integrated sequence-derived features (manual extraction, pre-trained language models) with structure-derived features (secondary structure contact maps).
• Applied GNNs to the entire protein sequence and its contact graph for comprehensive site prediction.

Main Results:

• GraphPhos demonstrated significant improvements in prediction accuracy for serine (≥8%), threonine (≥15%), and tyrosine (≥12%) sites.
• Achieved an average accuracy improvement of 7% compared to models predicting individual amino acid categories.
• The model effectively utilizes both sequence and structural information for enhanced prediction.

Conclusions:

• GraphPhos represents a powerful new tool for identifying protein phosphorylation sites.
• The integration of sequence and structure features within a GNN framework offers superior prediction performance.
• This approach advances research in protein phosphorylation and its associated biological functions.