Delineating the Spatial Patterns of Cell Type–Specific Stemness Using a Statistical Deconvolution Model

Summary Cellular stemness is a cells ability to self-renew and differentiate into specialized cell types. In spatial transcriptomics (ST), researchers can use gene activity to quantify stemness and the associated spatial locations, allowing for the investigation of cellular profiles in their natural context. However, many ST platforms provide data with low resolution, and each data spot (i.e., spatial location) contains multiple cell types, which hinders the investigation of cell type–specific (CTS) behaviors, such as cell stemness, at different spatial locations. We developed a bivariate kernel-weighted regression method with constrained optimization to estimate CTS stemness as a function of spatial location and developed accompanying visualization tools. Through simulation studies and an application to real breast cancer ST data from the 10x Genomics Visium platform, we demonstrated that our method can accurately estimate CTS stemness and help shed light on the interplay among cell type, tissue structure, and stemness. Availability and implementation An R Shiny app implementing the proposed method is available at GitHub ( https://github.com/ericli0480/stemness-deconvolution/ ). Supplementary information Available in a separate PDF file