Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST
Advances in spatial transcriptomics technologies have enabled the gene expression profiling of tissues while retaining its spatial context. Effective exploitation of this data combination requires spatially informed analysis tools to perform three key tasks, spatial clustering, multi-sample integration, and cell type deconvolution. Here, we present GraphST, a novel graph self-supervised contrastive learning method that incorporates spatial location information and gene expression profiles to accomplish all three tasks in a streamlined process while outperforming existing methods in each task. GraphST combines graph neural networks with self-supervised contrastive learning to learn informative and discriminative spot representations by minimizing the embedding distance between spatially adjacent spots and vice versa. With GraphST, we achieved 10% higher clustering accuracy on multiple datasets than competing methods, and better delineated the fine-grained structures in tissues such as the brain and embryo. Moreover, GraphST is the only method that can jointly analyze multiple tissue slices in both vertical and horizontal integration while correcting for batch effects. Lastly, compared to other methods, GraphST’s cell type deconvolution achieved higher accuracy on simulated data and better captured spatial niches such as the germinal centers of the lymph node in experimentally acquired data. We further showed that GraphST can recover the immune cell distribution in different regions of breast tumor tissue and reveal spatial niches with exhausted tumor infiltrating T cells. Through our examples, we demonstrated that GraphST is widely applicable to a broad range of tissue types and technology platforms. In summary, GraphST is a streamlined, user friendly and computationally efficient tool for characterizing tissue complexity and gaining biological insights into the spatial organization within tissues.
PDF Abstract
