Reconstructing multi-scale tissue spatial architecture from single-cell RNA-seq with REMAP.

Area of Science:

• Computational Biology
• Genomics
• Systems Biology

Background:

• Cellular spatial organization is crucial for understanding tissue function and disease.
• Single-cell RNA sequencing (scRNA-seq) provides high-resolution gene expression but lacks spatial context.
• Spatial transcriptomics (ST) retains spatial information but faces limitations in cost and gene coverage.

Purpose of the Study:

• To develop a deep learning framework (REMAP) for reconstructing multi-scale spatial organization of scRNA-seq data.
• To integrate gene expression with neighborhood-level gene-gene covariance for spatial reconstruction.
• To enable spatial hypothesis generation and microenvironment discovery from cost-efficient single-cell data.

Main Methods:

• Developed REMAP, a deep learning framework integrating gene expression and neighborhood gene-gene covariance.
• Utilized one or multiple ST references for spatial reconstruction.
• Applied REMAP to diverse datasets including 2D/3D mouse brain, human fetal cortex, and seven human cancer types.

Main Results:

• REMAP consistently outperformed existing methods across various tissue types and species.
• Successfully resolved microglial neighborhood heterogeneity and identified a rare pro-inflammatory microglia-astrocyte subpopulation in a multiple sclerosis atlas.
• Recovered conserved, prognostically significant cancer-associated fibroblast subtypes in diverse cancers.

Conclusions:

• REMAP effectively transforms scRNA-seq data into spatially interpretable tissue maps.
• The framework facilitates discovery of cellular neighborhoods and microenvironments in health and disease.
• REMAP enables population-scale inference of conserved and perturbed tissue architectural principles.