STALocator is a method that spatially localizes scRNA-seq data through integration with ST data. STALocator is a deep learning-based tool consisting of an integration network that integrates scRNA-seq data with ST data and a localization network that predicts spatial location for scRNA-seq data. Among them, the integration network adopted the modified domain translation networks equipped with sliced Wasserstein distance to robustly align scRNA-seq and ST data. The localization network adopted the spatial location-supervised auto-encoder equipped with well-designed loss function to robustly fit low-dimensional representations and spatial locations for ST data. We designed different pipelines for ST data with different resolutions, which can be applied on different biological scenarios.

For low-resolution ST data, such as Spatial Transcriptomics and 10x Visium, we first train an integration network to obtain a low-dimensional representation that removes batch effects, and then train a localization network to predict the spatial location of cells. Single-cell data with spatial location information can be can be considered as a higher-resolution form of spatial transcriptome data compared to raw ST data. For high-resolution ST data, such as Slide-seq and FISH, it already has a resolution close to that of scRNA-seq data, so we only train the integration network, and update the optimal transport (OT) plan during the training process, and finally we can obtain a global OT plan from scRNA-seq data to ST data which can be used to enhance ST data. For Slide-seq data, exhibits more distinct and discernible patterns than raw data. Furthermore, it is noteworthy that FISH data presents a limitation in terms of the number of measurable genes. In light of this constraint, the application of our approach enables the acquisition of genome-wide ST data, addressing this limitation effectively.

It is recommended to use a Python version between 3.7 and 3.9, and a R version above 4.2.

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

scanpy>=1.8.2,<=1.9.6
torch>=1.8.0,<=1.13.0
torchvision>=0.9.0,<=1.14.0
POT==0.9.0

In addition, if you choose to use GPU, the versions of torch and torchvision need to be compatible with the version of CUDA.

The important R packages used to process data and perform enrichment analysis are as follows:

Seurat==4.2
SeuratData==0.2
GSEABase==1.60
DOSE==3.24
fgsea==1.24
clusterProfiler==4.6
org.Mm.eg.db==3.16
AnnotationDbi==1.60

After download STALocator from Github, you can install STALocator via

cd STALocator-main
python setup.py build
python setup.py install

In addtion, if you choose to install STALocator in a virtual environment, you should install the imageio and igraph packages first.

And if you choose to use R in jupyter notebook, you should install the R kernel in jupyter notebook first.

The following are detailed tutorials. Some related additional files can be downloaded here.

1. Simulation experiment

2. Localization of human brain MTG and M1 cells on human DLPFC sections

3. Localization of human SCC cells on tissue sections

4. Data enhancement of mouse hippocampus Slide-seq dataset

5. Data extension of mouse visual cortex STARmap dataset