Learning spatial structures of proteins improves protein-protein interaction prediction.

Area of Science:

• Computational biology
• Bioinformatics
• Structural biology

Background:

• Protein structure is crucial for protein function and protein-protein interaction (PPI) prediction.
• Limited availability of experimentally determined protein structures hinders structure-based prediction.
• Predicted protein structures offer a viable alternative to enhance sequence-based prediction.

Purpose of the Study:

• To develop a novel end-to-end framework, TAGPPI, for predicting PPIs using only protein sequence information.
• To integrate spatial structure information from predicted protein contact maps into sequence-based PPI prediction.
• To establish a new benchmark for sequence-based PPI prediction by incorporating predicted structural topology.

Main Methods:

• TAGPPI employs 1D convolution on protein sequences to extract features.
• Graph learning is applied to contact maps generated by AlphaFold to capture spatial information.
• The framework integrates sequence-derived and structure-derived features for comprehensive analysis.

Main Results:

• TAGPPI significantly outperforms nine state-of-the-art sequence-based PPI prediction methods across all evaluated metrics.
• The inclusion of spatial information from predicted contact maps demonstrably enhances PPI prediction accuracy.
• This represents the first method utilizing predicted protein topology graphs for sequence-based PPI prediction.

Conclusions:

• TAGPPI offers a powerful and accurate approach for predicting protein-protein interactions using only sequence data.
• The framework's ability to leverage predicted structural information overcomes limitations of sparse experimental structure data.
• The proposed architecture holds potential for broader applications in other protein-related prediction tasks.