Spatially informed clustering, integration, and deconvolution of spatial transcriptomics with GraphST

Abstract

Spatial transcriptomics technologies generate gene expression profiles with spatial context, requiring spatially informed analysis tools for three key tasks, spatial clustering, multisample integration, and cell-type deconvolution. We present GraphST, a graph self-supervised contrastive learning method that fully exploits spatial transcriptomics data to outperform existing methods. It 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. We demonstrated GraphST on multiple tissue types and technology platforms. GraphST achieved 10% higher clustering accuracy and better delineated fine-grained tissue structures in brain and embryo tissues. GraphST is also the only method that can jointly analyze multiple tissue slices in vertical or horizontal integration while correcting batch effects. Lastly, GraphST demonstrated superior cell-type deconvolution to capture spatial niches like lymph node germinal centers and exhausted tumor infiltrating T cells in breast tumor tissue.

Single-cell and Spatial Transcriptomics Clustering with an Optimized Adaptive K-Nearest Neighbor Graph

“…Spatial transcriptomics is an emerging technology that provides a roadmap of transcriptional activity within tissue sections. To better decipher domains or cell types that are spatially coherent in both gene expression and histology, a number of integrative approaches to combine gene expression, spatial location, histology, and H&E image have been developed [38][39][40][41] . Integrating multi-modal information are expected to define cell types or domains accurately than using gene expression alone.…”

“…In the spatial transcriptomics data, we further compared aKNNO with four integrative approaches, BayesSpace 41 , GraphST 40 , SpaGCN 38 , and stLearn 39 . BayesSpace implements a full Bayesian model that uses the information from spatial neighborhoods for resolution enhancement for spatial clustering.…”

“…Recently, several integrative approaches combining expression, spatial location and histology images have been developed to improve the performance of spatial transcriptomics clustering [38][39][40][41] . We compared aKNNO with four integrative approaches, stLearn 39 , SpaGCN 38 , GraphST 40 , and BayesSpace 41 . We set their number of clusters to 31, matching the cluster number of aKNNO (Figs.…”

Spatially Resolved Gene Expression is Not Necessary for Identifying Spatial Domains

“…We conducted a comprehensive benchmarking study comparing BINARY with recent state-of-the-art spatial clustering methods, which included SpaGCN 15 , SEDR 25 , STAGATE 24 , SpaceFlow 12 , and GraphST 26 , across 6 Spatially Resolved Transcriptomics (SRT) technologies (Fig. 1B).…”

“…As done by previous spatial clustering methods 15,26 , after the initial clustering outcomes are obtained, an optional step of clustering result optimization can be applied to achieve more coherent and smoother cluster delineations. For any given cell or spot, its label is reassessed and potentially reallocated.…”