Reference-Informed Spatial Domain Detection Using Weak Supervision for Spatial Transcriptomics

Area of Science:

• Spatial transcriptomics
• Computational biology
• Bioinformatics

Background:

• Spatial transcriptomics (ST) enables mapping tissue organization and function.
• Accurate tissue segmentation is crucial for ST data analysis.
• Existing methods often require extensive manual annotation or lack generalizability.

Purpose of the Study:

• To develop a novel, reference-informed, weakly supervised contrastive learning model for tissue segmentation in ST studies.
• To integrate manual annotations (scribbles) with gene expression profiles for improved segmentation accuracy.
• To provide a user-friendly software toolkit for spatial domain detection.

Main Methods:

• Introduced GraphScrDom, a contrastive learning model.
• Integrated expert-provided scribbles on spatial grids/histology images.
• Utilized cell type-specific gene expression profiles from reference single-cell RNA-seq data.
• Developed an integrative software toolkit with annotation and training modules.

Main Results:

• GraphScrDom achieved superior performance in tissue segmentation across various ST platforms and resolutions (bulk and single-cell).
• The model demonstrated strong generalizability and robustness, outperforming existing methods with limited annotations.
• Performance was validated using six widely adopted metrics.

Conclusions:

• GraphScrDom offers a robust and efficient solution for spatial domain detection in ST data.
• The developed toolkit facilitates user-friendly spatial domain analysis.
• This approach enhances the mapping of complex tissue organization and function.