Multi-Sample and Multi-Group Spatial Colocalization Analysis Using PANORAMIC

Spatial omics technologies that map cellular organization in individual tissue samples at high resolution require statistical analysis of spatial patterns that accounts for the hierarchical relationship of within- and between-sample variability. Existing spatial statistical methods typically overlook within-sample variability, resulting in unreliable inference and reduced reproducibility when analyzing new groups of samples. We introduce PANORAMIC, a hierarchical framework for efficient estimation and statistical testing of colocalization between pairs of cell types in spatial omics. PANORAMIC integrates spatial bootstrapping to quantify within-sample variability with random-effects meta-analysis to pool information on the colocalization of each cell pair and account for heterogeneity across independent samples within a group. Simulation analyses of images of multiple cell types generated from spatial point process models show that PANORAMIC accurately estimates variance components at the sample and group levels and outperforms naive pooling in key benchmarks. Applied to spatial omics images from colorectal cancer tissue microarrays imaged using a high-multiplexed immunofluorescence platform, PANORAMIC identifies higher colocalization among B- and T-cell subsets in samples with Crohns-like reactions (CLR) compared with samples exhibiting diffuse inflammatory infiltration (DII). These findings are consistent with tertiary lymphoid structure formation, more favorable prognosis in CLR versus DII, and were missed by existing computational methods. By modeling uncertainty at multiple hierarchical levels and enabling properly powered between-group comparisons, PANORAMIC addresses a gap in spatial omics analysis. PANORAMIC is broadly applicable across single-cell spatial technologies and tissue contexts and is released as an open-source R package, supporting statistical analysis of multi-sample and multi-group spatial omics.