Building a single cell transcriptome-based coordinate system for cell ID with SURE

Abstract

The growing scale of single cell data demands standardized cell identification. We develop SURE, a data-driven method that establishes a cell coordinate system through optimally-sized metacells exhibiting enhanced transcriptional homogeneity. This approach enables precise cellular state characterization while providing three key atlas functionalities: zero-shot query-to-reference mapping, out-of-atlas detection, and cell type annotation. SUREs hierarchical assembly pipeline successfully integrates large-scale atlas datasets, as demonstrated by constructing the Human Blood Metacell Atlas (HBMCA) from 5 million cells. Our method addresses the fundamental challenge of consistent cell positioning across diverse experimental contexts, facilitating robust comparative analyses between healthy and disease states.