SpatialZoomer: multi-scale feature analysis of spatial transcriptomics

The Python implementation of SpatialZoomer, a scalable toolkit for multi-scale spatial feature analysis of single-cell resolution spatial transcriptomics via spectral graph signal processing.

1. Introduction
2. Demos
3. System Requirements
4. Installation
5. Tutorials
6. Cite

Introduction

SpatialZoomer is a toolkit that enables the multi-scale analysis of spatial transcriptomics data. By modeling gene expression as signals on spatial graphs and applying spectral graph signal processing, the toolkit efficiently extracts multi-scale spatial features and identifies biologically meaningful structures. It can automatically identify “critical” scales by partitioning the cross-scale similarity map.

Functions

1. Zoom-capable analysis across spatial scales
   Automatically identifies spatial structures ranging from cell, niche, to domain level across critical scales.

2. Identification of spatial context-dependent subtypes
   As spatial scale increases, cells incorporate broader spatial context and gradually resolve into subclusters influenced by their microenvironments.

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

Advantages

• High computational efficiency with low memory usage
• Fast runtime for high-resolution datasets
• Capable of processing > 1 million cells on desktops
• No GPU required

Demos

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

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

👉 Demo Website

System Requirements

SpatialZoomer is lightweight and can be executed on laptops, desktops or servers. We tested it on the four machines:

• Most datasets can be processed within 1 hour to generate multi-scale results (10+ scales).

• The Xenium Prime Ovarian Cancer dataset (>1.1M cells, 5K genes) exceeds memory limit on a 32 GB laptop.

• The Xenium Prime Cervical Cancer dataset (≈800K cells, 5K genes) runs successfully on a 32 GB machine within ~4 hours.

• Figure 1: Peak memory usage across devices (add image link here)

• Figure 2: Total runtime across datasets (add image link here)

This package requires only a standard computer with enough RAM to support the in-memory operations with GPU-free.

设备与运行结果: 32G台式机因为内存爆炸运行不了1,10+万细胞、5K基因的Xenium Prime Ovarian cancer样本，其余样本都可以运行，包括801,131细胞、5K基因的Xenium Prime Cervical cancer样本，能在4h内获得10+个重要尺度下的结果。其余机器均能快速运行所有样本。

在四个机器上的时间和内存（图片展示） 大部分样本能够在1h内获得10余个尺度的多尺度结果 图1（内存） 图2（时间），后续给地址

Installation

3.9以上环境 conda create -n spatialzoomer python=3.10 -y conda activate spatialzoomer conda install -c conda-forge pyarrow pip install SpatialZoomer==1.0.0

Requires Python ≥ 3.9

We recommend using conda:

# Create and activate virtual environment
conda create -n spatialzoomer python=3.10 -y
conda activate spatialzoomer

# Install dependencies
conda install -c conda-forge pyarrow

# Install SpatialZoomer
pip install SpatialZoomer==1.0.0

Tutorials

Tutorial 1: Multi-scale analysis 针对空间转录组数据，自动选取重要尺度，并提计算每个尺度下的空间结构 (github跳转地址).ipynb 以CosMx mouse brain为例 .ipynb 以Xenium V1 lung cancer为例 .ipynb 以Xenium V1 ovarian cancer 为例

Tutorial 2: Identify spatial context dependent clusters 随着尺度增大，细胞融入更多的context信息，逐渐分离为亚群 .ipynb 以Xenium V1 lung cancer为例 .ipynb 以Xenium V1 ovarian cancer 为例