Graph Contrastive Learning for Inferring Spatial Cell Composition from Integrated Single-cell RNA Sequencing and...

Area of Science:

• Computational Biology
• Genomics
• Bioinformatics

Background:

• Spatial transcriptomics (ST) technologies offer high-throughput profiling with spatial organization but often lack single-cell resolution.
• This limitation hinders the identification of cell-type-specific spatial patterns and gene expression variations.

Purpose of the Study:

• To develop a novel computational method, Graph Contrastive learning for Inferring spatial cell composition (GCIRS), for accurate cell-type deconvolution in spatial transcriptomics.
• To enhance the representation of spatial spots by integrating single-cell RNA sequencing (scRNA-seq) data.

Main Methods:

• GCIRS constructs spot-cell heterogeneous graphs and utilizes metapath-based reasoning to infer inter-spot connections.
• It builds spot-spot homogeneous graphs to preserve cell-type-specific information.
• A structure-aware graph contrastive learning framework aligns structural patterns between real and pseudo-spot graphs for cross-modality knowledge transfer.

Main Results:

• GCIRS significantly improves spatial cell composition inference accuracy compared to eleven state-of-the-art methods.
• The method excels in reconstructing cell distribution patterns across multiple datasets.
• GCIRS successfully revealed tumor heterogeneity in pancreatic cancer and delineated complex tissue structures in the mouse brain.

Conclusions:

• GCIRS provides a robust framework for accurate cell-type deconvolution in spatial transcriptomics.
• The method effectively integrates scRNA-seq and ST data to enhance spatial pattern analysis.
• GCIRS has broad applications in understanding tissue heterogeneity and complex biological structures.