Next generation statistical framework for next generation spatial transcriptomics data

The rapid advancement of spatial transcriptomic technologies, particularly in-situ hybridization based methods, has enabled the profiling of thousands of gene at subcellular resolution across large tissue sections. Commercial platforms such as Xenium and CosMx now routinely generate high-quality datasets of increasing size and complexity. However, existing analytical approaches, often adapted from single-cell genomics, fall short in addressing the specific challenges posed by spatial data, especially at scale. In this work, we present TranspaceR, a new R package that introduces computational and statistical methods tailored to the analysis of next-generation spatial transcriptomic datasets. Our framework includes novel quality control procedures, scalable gene selection strategies (especially for spatially variable genes), and dimensionality reduction techniques. These methods are based on an in-depth statistical characterization of spatial data. We demonstrate that the framework TranspaceR leads to a significatively faster selection of spatially variable genes compared to current methods. We also demonstrate how single-cell annotation tools can be leveraged for automated cell-type labeling within spatial contexts. Together, the tools in TranspaceR enable the efficient and robust analysis of imaging-based spatial transcriptomics data, providing a comprehensive statistical framework for next-generation spatial transcriptomics data analysis.