Tu206 - A Multi-tiered Statistical Framework for Characterizing Morphological Cell-to-cell Interactions in Spatial Transcriptomics

Abstract Text: In spatial transcriptomics, cell-to-cell interactions are critically important because they influence various biological processes and phenomena that are central to understanding tissue function and disease. Single-cell-level profiling in spatial transcriptomics sheds light on the heterogeneity and dynamics of individual cells. This study aims to establish a statistical pipeline for identifying regions of unusual activity in cell frequency and characterizing morphological cell-to-cell interactions in those regions, which are vital for immune responses. We begin by annotating cells based on their gene expression profiles and selecting two cell types of interest for further analysis. The spatial domain is divided into subdomains based on a biological length scale, defining a meaningful size for each subdomain (i.e., niches). In each subdomain, we apply a first-order statistic to assess the frequency of each cell type and identify regions with abnormal cell-type distribution. These subdomains are further analyzed using a second-order statistic to examine cell-to-cell spatial distances, potentially indicating local signaling interactions between the selected cell types. If unusual spatial interactions are detected, we use a third-order statistic to explore detailed collocational alignments between the two cell types. This multi-tiered approach provides a comprehensive view of how cell interactions are spatially organized and how localized environments influence tissue behavior. Ultimately, this method has the potential to uncover insights into tissue development, disease mechanisms, and therapeutic strategies by focusing on the spatial organization of cellular interactions. This framework is applied to spatial data from psoriasis skin biopsies in the PAUSE trial at the Immune Tolerance Network (ITN).