NOLAN: SELF-SUPERVISED FRAMEWORK FOR MAPPING CONTINUOUS TISSUE ORGANIZATION

Spatial transcriptomics offers unprecedented insights into tissue organization, yet current methods often overlook transitional zones between cellular niches. We introduce NOLAN, a self-supervised framework that goes beyond detecting discrete niches to capture the continuous spectrum of tissue organization patterns. NOLAN learns cell representations informed by their neighborhoods, capturing variation within niches and across their boundaries. Using these representations, NOLAN constructs a graph-based abstraction of the tissue, modeling it as a network of interconnected regions bridged by transitional zones. Applying NOLAN to a multi-cancer spatial transcriptomics atlas, we uncover a landscape of both tissue-specific and shared cellular niches. Crucially, NOLAN reveals the continuous gradients of gene expression and cell type composition across these transitional zones, showcasing the ability of NOLAN to build a common coordinate system of tissues in an integrative analysis.