Abstract
Advances in spatial transcriptomics technologies has enabled gene expression profiling of tissues while retaining the spatial context. To effectively exploit the data, spatially informed analysis tools are required. Here, we present DeepST, a versatile graph contrastive self-supervised learning framework that incorporates spatial location information and gene expression profiles to accomplish three key tasks, spatial clustering, spatial transcriptomics (ST) data integration, and single-cell RNA-seq (scRNA-seq) transfer onto ST. DeepST combines graph neural networks (GNNs) with contrastive self-supervised learning to learn spot representations in the ST data, and an auto-encoder to extract informative features in the scRNA-seq data. Spatial contrastive self-supervised learning enables the learned spatial spot representation to be more informative and discriminative by minimizing the embedding distance between spatially adjacent spots and vice versa. With DeepST, we found biologically consistent clusters with greater accuracy than competing methods. We next demonstrated DeepST’s ability to jointly analyze multiple tissue slices in both vertical and horizontal integration while correcting for batch effects. Lastly, we used DeepST to deconvolute cell types present in ST with scRNA-seq data, showing better performance than cell2location. We also demonstrated DeepST’s accurate cell type mapping to recover immune cell distribution in the different regions of breast tumor tissue. DeepST is an easily usable and computationally efficient tool for capturing and dissecting the heterogeneity within ST data, enabling biologists to gain insights into the cellular states within tissues.
Competing Interest Statement
The authors have declared no competing interest.