PPI-Graphomer: enhanced protein-protein affinity prediction using pretrained and graph transformer models

Area of Science:

• Computational biology
• Structural bioinformatics
• Molecular modeling

Background:

• Protein-protein interactions (PPIs) are fundamental to biological processes.
• Understanding PPIs is crucial for biological mechanism elucidation and drug discovery.
• Protein binding interfaces, especially hotspot residues, are key to PPIs.
• Current deep learning models often fail to fully capture interface information.

Purpose of the Study:

• To develop an advanced computational module for enhanced characterization of protein binding interfaces.
• To improve the accuracy of predicting and analyzing protein-protein interactions.
• To overcome limitations of existing deep learning methods in modeling protein interfaces.

Main Methods:

• Proposed the PPI-Graphomer module, integrating pretrained features from large-scale language models and inverse folding models.
• Defined edge relationships and interface masks based on molecular interaction data.
• Utilized benchmark datasets for model training and evaluation.

Main Results:

• The PPI-Graphomer module demonstrated superior performance compared to existing methods.
• The model achieved high accuracy across multiple benchmark datasets.
• Strong generalization capabilities were observed, indicating robustness.

Conclusions:

• The PPI-Graphomer module significantly enhances the characterization of protein binding interfaces.
• This approach offers a powerful tool for studying protein-protein interactions and their biological significance.
• The findings have implications for advancing drug research and development.