N-Orbit: towards a universal model and metric for comparing tissue microenvironments

Abstract

Advances in spatial omics technologies have greatly expanded our ability to explore tissue microenvironments (TMEs) across development and disease. Yet a theoretical framework for modeling and comparing tissue architecture in diverse biological contexts remains lacking. We introduce, N-Orbit, a mathematical model that captures both cell type composition and spatial relationships within tissue cellular neighborhoods, encoding this information in a vector format for computationally efficient distance metric calculations. While not a neighborhood detection method itself, N-Orbit enhances insights gleaned from the neighborhood generated by the plethora of recently developed methods. We benchmarked the N-Orbit-based neighborhood distance metric on spatial omics datasets containing ground truth neighborhoods and patient clinical outcomes. We demonstrated that the N-Orbit outperforms cell type enrichment (CTE)-based metrics in discriminating neighborhood types, predicting clinical variables, and identifying homologous structures across species. Additionally, N-Orbit enhances model interpretability by allowing neighborhoods to be traced back to their enriched N-Orbit structures. N-Orbit holds significant potential for deepening our understanding of how TMEs remodel during development, disease, and evolution.