Spatial transcriptomics (ST) enables gene expression profiling while preserving the spatial architecture of intact tissue. Analyzing ST data often proceeds by first extracting cell-level information, typically through cell segmentation or cell-type deconvolution. However, a critical oversight has been that a substantial portion of molecular expression is systematically lost or unannotated by these methods. This lost expression can arise from diverse and biologically important sources like fragile or underrepresented cell types, subcellular structures like neurites, and extracellular expression. These omissions can result in biased analyses and incorrect or incomplete biological interpretations. We describe a new computational method, RESCUE, that can recover the unattributed spatial expression patterns missed by existing ST analysis methods and enable robust inference even when reference is incomplete. We validate RESCUE using MERFISH data from the honey bee brain and apply it to multiple ST datasets to demonstrate how it can reveal novel insights into complex tissue biology.
Lee et al. (Fri,) studied this question.