Eliason J, Rao A, Frankel TL, Peruzzi M (2026) Joint Modeling of Spatial Dependencies Across Multiple Subjects in Multiplexed Tissue Imaging. Annals of Applied Statistics (accepted)
https://arxiv.org/abs/2504.02693
Multiplexed imaging gives you detailed spatial maps of cells in tissue, but most methods analyze each tissue sample on its own. We build a Bayesian hierarchical model that pools information across subjects to learn how different cell types are spatially arranged relative to each other… which types cluster together, which repel, and how these patterns vary.
Under the hood, cell-type intensities are modeled as a multivariate log-Gaussian Cox process driven by a latent Gaussian process. The hierarchical setup lets you estimate spatial correlation functions across subjects while still capturing subject-level variability. It works well in simulations and on real imaging datasets, where it picks up distinct spatial organization patterns across disease subtypes.
