Inference of cell-type composition and single-cell spatial maps from spatial transcriptomics data with SWOT

Abstract

Spatially resolved single-cell transcriptomics is crucial for mapping the cellular atlas of organisms, but many spatial transcriptomics data lack single-cell resolution. Most cell-type deconvolution methods are limited to estimating cell-type proportions, and they cannot further identify the exact cells needed to reconstruct a single-cell spatial map. To overcome this limitation, we introduce a spatially weighted optimal transport method, named SWOT, for learning a mapping from cells to spots to infer both cell-type composition and single-cell spatial maps from spot-based spatial transcriptomics data. Experimental results demonstrate that the learned cell-to-spot mapping offers advantages in estimating cell-type proportions, cell numbers per spot, and spatial coordinates per cell. SWOT also depicts cell-type spatial distributions and maps single cells to their spatial locations in different morphological tissues. We further showcase the utility of SWOT in assistance of accurately identifying and functionally annotating cellular neighborhoods for deciphering tissue architecture. In summary, SWOT represents a useful tool for transforming abundant spot-resolution spatial transcriptomics data into single-cell resolution, thereby facilitating cell-level discoveries within tissues.

SWOT infers cell-type composition and single-cell spatial maps from spatial transcriptomics data, transforms abundant spot-resolution data into single-cell resolution, and promotes cell-level discoveries within tissues.