EGPDI: identifying protein-DNA binding sites based on multi-view graph embedding fusion

Area of Science:

• Computational biology
• Bioinformatics
• Genomics

Background:

• Protein-DNA interactions are vital for biological processes.
• Accurate prediction of these binding sites is crucial for genetic analysis, protein function studies, and drug discovery.
• Existing computational methods have limitations due to reliance on handcrafted features and single-model architectures.

Purpose of the Study:

• To develop a novel computational method, EGPDI, for accurate protein-DNA binding site prediction.
• To address the limitations of current methods by employing a multi-view graph embedding fusion approach.
• To improve the analysis of genetic material, protein functions, and drug design.

Main Methods:

• Integration of Equivariant Graph Neural Networks (EGNN) and Graph Convolutional Networks II (GCNII) for global and local node embedding extraction.
• Utilization of a gated multi-head attention mechanism for effective fusion of dual embedding representations.
• Incorporation of additional node features from protein language models to enhance structural information.

Main Results:

• EGPDI demonstrates superior performance compared to state-of-the-art methods in protein-DNA binding site prediction.
• Five-fold cross-validation and independent testing confirm the method's accuracy and effectiveness.
• Comparative experiments and case studies validate the superiority and generalization ability of EGPDI.

Conclusions:

• EGPDI represents a significant advancement in computational protein-DNA binding site prediction.
• The multi-view graph embedding fusion strategy effectively captures complex interaction patterns.
• The method offers a powerful tool for biological research and therapeutic development.