Robust characterization and interpretation of rare pathogenic cell populations from spatial omics using GARDEN

Area of Science:

• Computational biology
• Spatial omics
• Pathology

Background:

• Spatial omics provides tissue context for molecular data.
• Current tools often miss rare cell populations crucial for disease.
• Identifying these rare cells is vital for understanding disease mechanisms.

Purpose of the Study:

• To develop a computational framework, GARDEN, for identifying and characterizing rare pathogenic cells in spatial omics data.
• To improve the detection of disease-driving cell populations.
• To enable 3D reconstruction of tissue anatomy from spatial omics data.

Main Methods:

• GARDEN utilizes a spatially-aware graph fusion contrastive model with graph-based dynamic attention.
• The framework is designed for consistency across tissues, species, and resolutions.
• It includes an interpretation module for pinpointing immune signatures and extends to spatial chromatin accessibility.

Main Results:

• GARDEN successfully identified pathogenic cells in Alzheimer's disease, nasopharyngeal carcinoma, and breast cancer models.
• It localized specific microglia populations in Alzheimer's disease and identified tertiary lymphoid structures in nasopharyngeal carcinoma.
• The framework uncovered inflammatory macrophages linked to pro-metastatic signaling in breast cancer and provided insights into epigenetic regulation.

Conclusions:

• GARDEN is a robust computational framework for discovering rare pathogenic cells in spatial omics.
• It enhances understanding of disease mechanisms and cellular interactions.
• The tool has broad applications in diagnostics and therapeutic targeting, including epigenetic insights.