SHADE: A Multilevel Bayesian Approach to Modeling Directional Spatial Associations in Tissues.

Motivation: Spatial dependencies in tissue microenvironments, particularly asymmetric interactions between cell types, are central to understanding immune dynamics, tumor behavior, and tissue organization. Existing spatial statistical methods often assume symmetric associations or analyze images independently, limiting biological interpretability and inference quality. Results: We introduce SHADE (Spatial Hierarchical Asymmetry via Directional Estimation), a Bayesian hierarchical framework that models asymmetric spatial associations and multilevel structure in multiplexed imaging data. SHADE captures directional relationships via smooth spatial interaction curves (SICs), provides interpretable distance-resolved summaries of cell-cell interactions, and supports multiscale inference across tissue sections, patients, and cohorts. Simulation studies demonstrate improved inference quality and robustness, and application to colorectal cancer imaging data reveals biologically meaningful differences in immune and stromal organization. Availability and Implementation: Source code and analysis scripts are freely available at http://github.com/jeliason/SHADE and http://github.com/jeliason/shade_paper_code, implemented in R and Stan.