scNET: learning context-specific gene and cell embeddings by integrating single-cell gene expression data with...

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Single-cell RNA sequencing (scRNA-seq) reveals tissue heterogeneity but struggles with pathway and complex identification due to data limitations.
• scRNA-seq data is characterized by high noise and zero inflation, complicating accurate biological interpretation.
• Protein-level interactions are crucial for understanding cellular pathways and complexes, complementing gene expression data.

Purpose of the Study:

• To develop a novel computational approach for integrating scRNA-seq data with protein-protein interaction networks.
• To address the limitations of gene expression data in capturing cellular pathways and complexes.
• To improve the analysis of scRNA-seq data by leveraging network information for enhanced biological insights.

Main Methods:

• A dual-view graph neural network architecture (scNET) was developed for joint representation learning.
• The method integrates gene expression profiles with protein-protein interaction networks.
• An attention mechanism was employed to refine cell-cell relationships and model context-specific gene-to-gene interactions.

Main Results:

• scNET demonstrated superior performance in gene annotation and pathway characterization.
• The approach effectively identified gene-gene relationships within biological contexts.
• Evaluations showed significant improvements in cell clustering and pathway analysis across diverse cell types and conditions.

Conclusions:

• Integrating scRNA-seq with protein-protein interaction networks via graph neural networks offers a powerful approach to decipher cellular heterogeneity.
• scNET enhances the biological interpretability of scRNA-seq data by capturing pathway and complex dynamics.
• This method provides a robust framework for advancing single-cell data analysis in various biological contexts.