PPAP: A Protein-protein Affinity Predictor Incorporating Interfacial Contact-Aware Attention.

Area of Science:

• Computational Biology
• Structural Biology
• Machine Learning

Background:

• Protein-protein interactions (PPIs) are crucial for biological functions and drug discovery.
• Accurate prediction of PPI binding affinity is essential for understanding molecular mechanisms and protein engineering.
• Recent advances in protein structure prediction offer new avenues for structure-based affinity prediction.

Purpose of the Study:

• To develop a novel deep learning framework, PPAP, for enhanced prediction of protein-protein interaction binding affinity.
• To integrate structural information with sequence-based features for more accurate affinity prediction.
• To improve upon existing methods that primarily rely on sequence data.

Main Methods:

• Developed PPAP, a deep learning framework integrating structural and sequence features.
• Employed an interfacial contact-aware attention mechanism to capture interaction interface details.
• Evaluated model performance on internal and external test datasets.

Main Results:

• PPAP demonstrated superior prediction performance compared to sequence-based methods.
• Achieved a Pearson correlation coefficient (R) of 0.540 and MAE of 1.546 on the internal test set.
• Outperformed benchmarked models on the external test set with a higher R of 0.63.
• Incorporating PPAP predictions enhanced protein binder design enrichment by up to 10-fold.

Conclusions:

• PPAP offers a robust and accurate approach for predicting protein-protein interaction binding affinity.
• The framework effectively utilizes structural insights from protein complexes.
• PPAP shows significant potential for applications in protein design and other protein interaction-related research.