Accurate Protein-Protein Interaction Prediction: Based on Multiview Heterogeneous Graph Autoencoders and Random...

Area of Science:

• Computational Biology
• Biochemistry
• Bioinformatics

Background:

• Protein-protein interactions (PPI) are crucial for cellular functions and drug discovery.
• Current deep learning models for PPI prediction are limited by their reliance on sequence data and poor integration of structural features.

Purpose of the Study:

• To develop a novel model, MEGAE, for high-precision prediction of protein-protein interactions (PPI) and protein-protein interaction sites (PPIS).
• To overcome limitations of existing methods by integrating diverse protein data, including sequence, structure, and physicochemical properties.

Main Methods:

• MEGAE reconstructs amino acid microenvironments using a vector quantization autoencoder, fusing physicochemical, structural, and sequence data.
• A multiview random masking strategy enhances the robustness of microenvironment embeddings.
• Graph neural networks (GNNs) are employed with protein graphs and interaction networks to capture multilevel relationships.

Main Results:

• MEGAE achieves high-precision prediction of PPI and PPIS.
• The model outperforms state-of-the-art sequence- and structure-based methods across multiple datasets.
• Demonstrates higher accuracy in predicting both interaction types and specific interaction sites.

Conclusions:

• MEGAE represents a significant advancement in PPI and PPIS prediction through microenvironment-aware modeling.
• The integrated approach enhances the understanding of complex protein interactions.
• This method holds promise for accelerating targeted drug development and elucidating cellular mechanisms.