ProAffinity-GNN: A Novel Approach to Structure-Based Protein-Protein Binding Affinity Prediction via a Curated Data...

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Protein-protein interactions (PPIs) are fundamental to biological processes and disease.
• Accurate determination of binding affinities is crucial but challenging due to high costs of experimental and computational methods.
• Existing quantitative binding affinity prediction methods lack comprehensive data.

Purpose of the Study:

• To address the need for effective binding affinity prediction methodologies.
• To create the largest dataset for structure-based pairwise protein interaction with binding affinity.
• To develop and validate a novel deep learning framework for predicting protein-protein binding affinity.

Main Methods:

• Manually curated pairwise interaction labels from 3D structures of protein complexes with experimentally determined binding affinities.
• Developed ProAffinity-GNN, a deep learning framework integrating protein language models and graph neural networks (GNNs).
• Evaluated model performance on benchmark datasets and a case study.

Main Results:

• Created the largest dataset for structure-based protein-protein binding affinity prediction.
• ProAffinity-GNN demonstrated superior accuracy compared to existing models.
• The model exhibited strong generalization capabilities across diverse datasets.

Conclusions:

• ProAffinity-GNN offers a significant advancement in predicting structure-based protein-protein binding affinities.
• The developed framework and dataset facilitate further research in protein engineering and drug discovery.
• This approach provides a more efficient and accurate alternative to traditional methods for binding affinity determination.