Protein-protein interaction prediction using bidirectional GRUs with explicit ensemble.

Area of Science:

• Computational Biology
• Bioinformatics
• Molecular Biology

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular functions across all organisms.
• Accurate in-silico identification of PPIs is vital for protein function prediction and drug design.
• Current sequence-based PPI prediction models often lack comprehensive sequence characterization, limiting cross-species applicability.

Purpose of the Study:

• To develop a more comprehensive method for characterizing protein sequences for PPI prediction.
• To improve the accuracy and generalizability of in-silico PPI identification across different species.

Main Methods:

• Utilized the SVHEHS descriptor combined with advanced feature coding techniques for comprehensive protein sequence characterization.
• Employed bidirectional gated recurrent units for multi-information fusion.
• Evaluated the model on H. pylori and S. cerevisiae datasets.

Main Results:

• Achieved high prediction accuracies of 96.47% (H. pylori) and 97.79% (S. cerevisiae).
• Outperformed most existing state-of-the-art models in PPI prediction.
• Demonstrated strong generalizability across diverse species datasets.

Conclusions:

• The proposed model offers a more robust approach to in-silico PPI prediction.
• The enhanced sequence characterization and fusion techniques improve cross-species prediction performance.
• This model can serve as a valuable tool for studying protein interaction networks in various species.