Year: 2026
Type: article
Abstract: High-resolution three-dimensional (3D) tissue atlases promise to link cellular architecture to tissue function, but assembling them from serial sections containing millions of cells remains challenging. Here we present Space-map, an accurate and efficient method for reconstructing 3D tissue models from high-content spatial molecular profiles. Space-map combines multiscale feature matching with large deformation diffeomorphic metric mapping to automate global alignment across serial sections while preserving local microanatomy. We generate spatial transcriptomics (Xenium; ~2.7 million cells) and proteomics (CODEX; ~4.8 million cells) datasets from serial sections to build 3D models of a colon polyp and healthy intestinal tissues. These models reveal cell-type heterogeneity in the polyp and preserved crypt and glandular architecture and mucosal-cavity continuity in healthy intestines. Benchmarking on our own and publicly available datasets demonstrates that Space-map achieves substantially higher registration accuracy and up to a tenfold speedup over existing methods, supporting its use as a robust framework for constructing 3D human tissue atlases at single-cell resolution.
Source: Nature Methods
Institutions: Santa Clara University, Nankai University, Stanford Medicine, Stanford University, Sun Yat-sen University, Sun Yat-sen Memorial Hospital, Washington University in St. Louis, University of Michigan, Michigan Medicine, Stanford Cardiovascular Institute, Cardiovascular Institute of the South
Language: English
Subfield: Molecular Biology
Domain: Life Sciences
Open Access status: hybrid
Estimated APC paid: $12,690
OpenAlex ID: W7236855243
