SpaceBender: Denoising Spatial Transcriptomics Data to Enhance Biological Signals

Abstract Spatial transcriptomics (ST) allows for the simultaneous profiling of cell phenotype (e.g. transcriptome) and physical position. Although ST data has brought about numerous new biological insights, it remains limited by noise, largely in the form of RNA diffusion. Here, we introduce SpaceBender which leverages spatial-specific information (e.g. spatial ambient RNA niches) to build upon single-cell denoising strategies. SpaceBender outperforms current ST denoising methods in simulations and in vivo chimeric tissues. Through case studies, we demonstrate how SpaceBender unveils hidden biological insights and increases the significance of said insights as evaluated by statistical testing. Finally, we reveal how SpaceBender may also be applied to subcellular resolution data where it removes off-target expression of neighboring cell type specific marker genes. In all, we present SpaceBender as an ST denoising method, freely available as an open-source package, that may enhance the insights the field may draw from various ST data types.