Accurately Deciphering Tissue Heterogeneity From Spatial Multi-Modal and Multi-Omics With STransformer

Area of Science:

• Computational Biology
• Bioinformatics
• Deep Learning

Background:

• Spatially resolved technologies allow simultaneous data acquisition within tissue slices, preserving spatial context.
• Existing computational methods struggle to universally process diverse spatial multi-modal and multi-omics data.
• Understanding tissue heterogeneity is crucial for deciphering disease pathogenesis.

Purpose of the Study:

• To introduce STransformer, a unified deep learning framework for analyzing spatial multi-modal and multi-omics data.
• To develop a flexible computational approach for dissecting complex tissue heterogeneity.

Main Methods:

• Developed STransformer, a deep learning framework capturing short-range cellular interactions and tissue-wide patterns.
• Applied STransformer to diverse spatial data, including multi-modal and multi-omics datasets.
• Evaluated STransformer's performance across different species, tissue types, and data modalities.

Main Results:

• STransformer successfully delineated anatomical structures in the human cortex and uncovered mechanisms in Alzheimer's disease.
• Characterized spatiotemporal developmental trajectories in chicken cardiogenesis using spatial multi-modal data.
• Deciphered immune microenvironments in human tonsil and inferred regulatory mechanisms in mouse embryonic brain using spatial multi-omics data.

Conclusions:

• STransformer is a versatile and robust framework for analyzing complex spatial biological data.
• The framework advances the understanding of tissue heterogeneity and disease pathogenesis.
• Enables seamless processing of spatial multi-modal and multi-omics data for comprehensive biological insights.