Scale-resolved correlation functions unify spatial omics across platforms and resolutions

Abstract

Spatial omics platforms populate a multidimensional parameter space and differ in resolution, panel size, sensitivity, and modality. This diversity has led to platform-specific analyses whose results are rarely comparable. Here we present SCOTE (Spatial COvariance TEnsors), a computational framework that quantifies spatial relationships of all feature pairs across all length scales, for molecule coordinates and binned counts alike. Measuring the same quantity, with statistical errors, across platforms, samples, and patients, SCOTE separates measurement effects from biology, with platform artifacts appearing as identifiable, scale-localized features. SCOTE's molecule-level functions resolved a knockout effect in mouse spleen that cell-level analyses missed, and could show that tissue area and replicates, not molecule depth, set the statistical error. To match the growth of spatial omics data sets, SCOTE runs efficiently on CPUs and GPUs, and processes a billion-molecule whole-transcriptome section at all length scales, with uncertainties, in 26 minutes on a single GPU.