Abstract
The rapid development of spatial transcriptomics (ST) technologies has enabled transcriptome-wide profiling of gene expression in tissue sections. Despite the emergence of single-cell resolution platforms, most ST sequencing studies still operate at a multicell resolution. Consequently, deconvolution of cell identities within the spatial spots has become imperative for characterizing cell-type-specific spatial organization. To this end, we developed Spatial Deconvolution Explorer (SpatialDeX), a regression model-based method for estimating cell-type proportions in tumor ST spots. SpatialDeX exhibited comparable performance to reference-based methods and outperformed other reference-free methods with simulated ST data. Using experimental ST data, SpatialDeX demonstrated superior performance compared with both reference-based and reference-free approaches. Additionally, a pan-cancer clustering analysis on tumor spots identified by SpatialDeX unveiled distinct tumor progression mechanisms both within and across diverse cancer types. Overall, SpatialDeX is a valuable tool for unraveling the spatial cellular organization of tissues from ST data without requiring single-cell RNA-seq references. Significance: The development of a reference-free method for deconvolving the identity of cells in spatial transcriptomics datasets enables exploration of tumor architecture to gain deeper insights into the dynamics of the tumor microenvironment.