PCLT-PPI: Predicting Multi-Type Interactions Between Proteins Based on Point Cloud Structure and Local Topology...

Area of Science:

• Biochemistry
• Computational Biology
• Structural Biology

Background:

• Protein-protein interactions (PPIs) are fundamental to cellular processes.
• Current computational methods often overlook the impact of protein 3D structure and topology on PPIs.
• Neural network embeddings can distort protein topological relationships.

Purpose of the Study:

• To develop a novel computational method for predicting PPIs that incorporates protein 3D structure and preserves local topology.
• To improve the accuracy, robustness, and generalization ability of PPI prediction models.

Main Methods:

• Utilizing protein point cloud structures to extract spatial features.
• Employing graph neural networks to capture relational features between proteins.
• Implementing a local topology preservation mechanism in both original and embedding spaces (PCLT-PPI).

Main Results:

• PCLT-PPI significantly outperforms existing state-of-the-art PPI prediction methods across various test partitions and evaluation metrics.
• The method demonstrates superior performance in predicting unseen protein-protein interactions, indicating enhanced robustness and generalization.
• Incorporating point cloud structure and local topology preservation demonstrably improves PPI prediction.

Conclusions:

• PCLT-PPI offers a more effective approach to PPI prediction by leveraging 3D structural information and topology preservation.
• The findings highlight the importance of considering native spatial structures in computational biology.
• This work provides a valuable reference for future research in protein interaction prediction and structural bioinformatics.