STAMapper

Overview

STAMapper is a method that annotates cells from single-cell spatial transcriptomics (scST) data. It is a deep learning-based tool that uses a heterogeneous graph neural network to transfer the cell-type labels from single-cell RNA-seq (scRNA-seq) data to scST data.

We also collected 81 scST datasets consisting of 344 slices and 16 paired scRNA-seq datasets from eight technologies and five tissues, served as a benchmark for scST annotation (can be downloaded from here).

Prerequisites

It is recommended to use a Python version 3.9.

• set up conda environment for STAMapper:

conda create -n STAMapper python==3.9

• install STAMapper from shell:

conda activate STAMapper

• the important Python packages used to run the model are as follows:

torch>=1.12.0,<=2.0.1
torchvision>=0.13.0,<=0.15.2
dgl>=1.1.2,<=2.1.0

STAMapper is test on GPU, the versions of torch, torchvision and dgl need to be compatible with the version of CUDA.

Installation

You can install STAMapper via:

git clone https://github.com/zhanglabtools/STAMapper.git
cd STAMapper
python setup.py build
python setup.py install

Tutorials

The following are detailed tutorials. All tutotials were ran on a 12600kf cpu and a 3060 12G gpu.

1. Cell-type annotation on scST data (with pre-annotated info).

2. Cell-type annotation on scST data (without pre-annotated info).

3. Reannotation on scST data (with pre-annotated info).

4. Unknown cell type detection.

5. Gene module extraction.