scAspa Quantifies Spatial Information in Single-Cell Atlases

Spatial transcriptomics has extended single-cell genomics into the spatial domain, making it possible to study the nuanced effects of tissue location and local cellular neighborhood on gene expression. As a cell's expression profile is a product of its microenvironment, single-cell atlases should encode latent spatial information, despite the spatial context being lost during tissue dissociation. Identifying this retained spatial signal provides an avenue to resolve nuanced cell states in existing atlases. However, how much of this information remains encoded in single-cell atlas representations has not been systematically assessed. To address this gap, we introduce scAspa, a framework that quantifies the spatial information preserved in cellular embeddings, specifically focusing on niche and local environment signals. Applying scAspa to the Human Lung Cell Atlas (HLCA) and the Pan-gastrointestinal atlas (pan-GI atlas) integrated with tissue-matched Xenium spatial data, we characterize previously missed, spatially-distinct cell states of alveolar fibroblasts and secretory cells, identify Xenium cell segmentation artifacts, and optimize reference atlas integration pipelines for high resolution cell state identification. Taken together, scAspa demonstrates how spatial information can be systematically combined with single-cell reference atlases to refine both single-cell and spatial analyses. We envision that this framework will enable deeply-annotated spatially informed cell atlases and more robust spatial-analysis workflows.