An End-to-End Framework for Scalable Spatial Proteomics Analysis

Foundation models offer a promising alternative to marker-by-marker analysis in spatial proteomics, but their practical use remains constrained by heterogeneous imaging platforms, diversified marker panels, data formats, and fragmented analysis workflows. We introduce CORAL, an end-to-end framework for applying spatial-omics foundation models reproducibly at cohort scale. The current iteration of CORAL ingests cyclic immunofluorescence, multispectral and ion-based acquisitions from a range of instruments and file formats, standardizing each into a self-contained OME-Zarr store, and provides tissue-, cell- and patch-level analysis units that experts can inspect and correct in-the-loop. A single patch-centric representation then supports analyses spanning single-cell phenotyping, tissue composition, spatial-domain discovery, patient stratification, treatment response and artifact detection. CORAL was designed to standardize every encoder to consume identical patches through a singular interface, thus enabling like-for-like comparison of foundation models possible. We use CORAL to compare seven patch encoders across seven fluorescence-based cohorts spanning lymphoma, breast, kidney and head and neck tissue