Finding spatially variable ligand-receptor interactions with functional support from downstream genes

Spatial transcriptomics has emerged as a groundbreaking tool for the study of intercellular ligand-receptor interactions (LRIs) that exhibit spatial variability. To identify spatially variable LRIs with activation evidence, we present SPIDER, which constructs cell-cell interaction interfaces constrained by cellular interaction capacity, and profiles and identifies spatially variable interaction (SVI) signals with support from downstream transcript factors via multiple probabilistic models. SPIDER demonstrates superior performance regarding accuracy, specificity, and spatial variance relative to existing methods. Experiments of simulations and real datasets in bulk and single-cell resolutions validate SPIDER-identified SVIs by spatial autocorrelation and correlation with downstream target genes, and reveal their consistency across multiple biological replicates. Particularly, distinct SVIs on mouse datasets reveal the potential in representing regional and inter-cell type interactions. SPIDER groups SVIs with similar spatial distributions into SVI patterns that are supported by strong Pearson correlations on spot annotations, generating interaction-based sub-clusters within cell-type regions, and deriving enriched pathways. Detecting spatial variance across cell-cell interactions remains computationally challenging. Here, the authors develop SPIDER, a statistical and machine learning tool to detect functionally-supported spatially variable ligand-receptor interactions from spatial transcriptomics data at bulk and single-cell resolutions.