SPIDER: constructing cell-type-specific protein-protein interaction networks

Area of Science:

• Computational biology
• Systems biology
• Bioinformatics

Background:

• Protein-protein interactions (PPIs) are crucial for cellular functions but are context-dependent, varying across cell types, time, and space.
• Existing PPI detection assays provide static networks under non-endogenous conditions, failing to capture dynamic cellular complexity.
• There is a critical need for computational methods to predict cell-type-specific PPI networks.

Purpose of the Study:

• To develop a novel computational method for predicting cell-type-specific protein-protein interaction (PPI) networks.
• To improve the accuracy and biological relevance of PPI network predictions by incorporating experimental data.
• To facilitate the identification of tissue-specific disease genes using predicted PPI networks.

Main Methods:

• Introduced SPIDER (Supervised Protein Interaction DEtectoR), a graph attention-based model.
• Utilized experimentally measured cell-type-specific networks to guide the supervised training of the model.
• Evaluated the model's performance on human and mouse experimental data.

Main Results:

• SPIDER significantly outperforms existing unsupervised approaches in predicting cell-type-specific PPI networks.
• The model demonstrates strong generalization capabilities, predicting networks for tissues lacking prior experimental PPI data.
• Predicted networks facilitated the identification of tissue-specific disease genes.

Conclusions:

• SPIDER offers a powerful, data-driven approach to model dynamic, cell-type-specific protein-protein interactions.
• The method advances the understanding of cellular complexity and provides a valuable tool for disease gene discovery.
• The developed code and data are publicly available for further research and application.