DeepST: identifying spatial domains in spatial transcriptomics by deep learning

Abstract

Recent advances in spatial transcriptomics (ST) have brought unprecedented opportunities to understand tissue organization and function in spatial context. However, it is still challenging to precisely dissect spatial domains with similar gene expression and histology in situ. Here, we present DeepST, an accurate and universal deep learning framework to identify spatial domains, which performs better than the existing state-of-the-art methods on benchmarking datasets of the human dorsolateral prefrontal cortex. Further testing on a breast cancer ST dataset, we showed that DeepST can dissect spatial domains in cancer tissue at a finer scale. Moreover, DeepST can achieve not only effective batch integration of ST data generated from multiple batches or different technologies, but also expandable capabilities for processing other spatial omics data. Together, our results demonstrate that DeepST has the exceptional capacity for identifying spatial domains, making it a desirable tool to gain novel insights from ST studies.

Optimizing Xenium In Situ data utility by quality assessment and best practice analysis workflows

“…Identifying the most reliable tools to define these domains is of general interest, although no independent comparison is yet available. Therefore, we benchmarked 4 domain finder algorithms (neighborhood-based 43 , Banksy 44 , DeepST 45 , SpaGCN) against the regions identified by expert manual annotation using the coronal P56 section from Allen Brain Atlas 46 (Figure 5D-E). A total of 36 domains were manually annotated, and thus each method was adjusted to predict a similar number of domains (35)(36)(37).…”

Spatially aware self-representation learning for tissue structure characterization and spatial functional genes identification

“…Other spatial clustering methods also have good spatial continuity in spatial domains, but some (e.g., Giotto and Vesalius) undetected the refined boundaries of certain domains, thus reducing their clustering performance. Additionally, we verified the universality of SpaSRL in quantitative or qualitative ways based on whether data annotation is provided (e.g., Slide-seqV2 [30], Seq-Scope [31], 4i and MIBI-TOF [32]) (Supplementary Figures 7-9). These benchmark tests demonstrated the superiority of SpaSRL at identifying spatial functional domains accounting for spatial coherence and biological difference.…”

“…More notably, SpaSRL succeeded to figure out the finer-grained layered organization (from the outer layer of Bergmann cells to the inner of oligodendrocytes) of cerebellum (Figures 6A-C). Based on the locally enhanced expression X , we calculated the LFCs of the marker genes from RCTD [32] to quantify the clustering performance of each method and high average LFC value obviously indicates high biological specificity between the identified clusters. As expected, these computationally obtained clusters appeared to be more biologically meaningful than random separations (Figure 6D, Wilcoxon signed-rank test, P < 10 −30 ).…”

“…Additionally, we verified the universality of SpaSRL in quantitative or qualitative ways based on whether data annotation is provided (e.g., Slide-seqV2 [30], Seq-Scope [31], 4i and MIBI-TOF [32]) (Supplementary Figures 7-9). These benchmark tests demonstrated the superiority of SpaSRL at identifying spatial functional domains accounting for spatial coherence and biological difference.…”

Mapping the topography of spatial gene expression with interpretable deep learning

“…Many methods focus on the identification of distinct spatial domains by partitioning tissues into subregions having large, discontinuous changes in gene expression, e.g. [168, 58, 32, 104, 81, 153, 76, 167, 53], but do not model continuous gene expression gradients within these regions. Several other methods instead test whether the expression of an individual gene varies spatially by fitting a function to the observed transcript counts at spatial locations [132, 130, 171, 18, 145].…”