SpatialZoomer: multi-scale feature analysis of spatial transcriptomics

The Python implementation of SpatialZoomer, a toolkit for multi-scale spatial feature analysis of single-cell resolved spatial transcriptomics via spectral graph signal processing. It scales to millions of cells with high computational efficiency and low hardware requirements.

Explore SpatialZoomer's multi-scale results by sliding the scale axis across four datasets on our interactive demo site:

👉 Demo Website

👉 Github Website

Introduction

SpatialZoomer is a novel computational method that enables the multi-scale feature analysis of single-cell resolved spatial transcriptomics like a zoom-capable microscope. By modeling gene expression as signals on spatial neighbor graph and applying spectral graph signal processing, the method efficiently extracts spatial features across multiple scales and reveals biologically meaningful structures. It can automatically identify “critical” scales by partitioning the cross-scale similarity map via dynamic programming.

Functions

• Zoom-capable analysis across spatial scales
  SpatialZoomer enables the detection of spatial structures from cells, niches, to domains across multiple scales. It can cpature biologically meaningful signals at specific scales, such as those related to disease progression in perturbation or longitudinal scenarios.

• Identification of spatially dependent subtypes
  As the spatial scale increases, SpatialZoomer progressively integrates spatial context from larger neighborhoods into each cell’s transcriptomic profile, potentially enabling cells from a single cluster at scale 0 to separate into spatially dependent subclusters at larger scales. These spatial contexts can amplify transcriptomic heterogeneity between biologically distinct subtypes.

• Uncovering complex tissue architecture
  The extracted multi-scale features can be used to dissect the complex spatial organization.

Advantages

• High computational efficiency with relatively low memory requirements (No GPU required).
• Fast runtime, even for large-scale datasets.
• Scalable to >1 million cells on desktops.

System Requirements

SpatialZoomer requires only a standard computer with sufficient RAM for in-memory operations and runs without the need for a GPU. It is lightweight and can be executed on laptops, desktops, or servers.

Cite

Xinqi Li#, Yuhan Fan#, Yue Han, Wenbo Guo, Jingmin Huang, Nan Yan, Zeyu Chen, Yanhong Wu, Yuxin Miao, Lin Hou, Xuegong Zhang, Zeyu Chen, Jin Gu*. SpatialZoomer: multi-scale feature analysis of spatial transcriptomics. bioRxiv (2025).