DESpace: spatially variable gene detection via differential expression testing of spatial clusters

Abstract

Motivation: Spatially resolved transcriptomics (SRT) enables scientists to investigate spatial context of mRNA abundance, including identifying spatially variable genes (SVGs), i.e. genes whose expression varies across the tissue. Although several methods have been proposed for this task, native SVG tools cannot jointly model biological replicates, or identify the key areas of the tissue affected by spatial variability.

Results: Here, we introduce DESpace, a framework, based on an original application of existing methods, to discover SVGs. In particular, our approach inputs all types of SRT data, summarizes spatial information via spatial clusters, and identifies spatially variable genes by performing differential gene expression testing between clusters. Furthermore, our framework can identify (and test) the main cluster of the tissue affected by spatial variability; this allows scientists to investigate spatial expression changes in specific areas of interest. Additionally, DESpace enables joint modeling of multiple samples (i.e. biological replicates); compared to inference based on individual samples, this approach increases statistical power, and targets SVGs with consistent spatial patterns across replicates. Overall, in our benchmarks, DESpace displays good true positive rates, controls for false positive and false discovery rates, and is computationally efficient.

Availability and implementation: DESpace is freely distributed as a Bioconductor R package at https://bioconductor.org/packages/DESpace.

Figure 1.

Spatial clusters in three samples from LIBD (Maynard et al. 2021) (Visium), melanoma (Thrane et al. 2018) (spatial transcriptomics (Ståhl et al. 2016)) and mouse cerebellum (Cable et al. 2022) (Slide-seqV2 (Stickels et al. 2021)) datasets. Spatial clusters were obtained: via manual annotations form a pathologist (LIBD), BayesSpace (Zhao et al. 2021) (melanoma), and StLearn (Pham et al. 2023) (mouse cerebellum).

Figure 2.

Three examples of simulated SVGs, from the LIBD data, following bottom/right, circular, and annotations patterns. Examples of SVGs from mixture and inverted mixture patterns are presented in Supplementary Fig. S1.

Figure 3.

TPR verus FDR for SVG detections in the individual sample simulations. Rows and columns refer to the anchor data used in the simulation, and to the SV profiles, respectively. BayesSpace_DESpace, BayesSpace_findMarkers, and BayesSpace_FindAllMarkers, as well as their counterparts StLearn_DESpace, StLearn_findMarkers, and StLearn_FindAllMarkers, indicate DESpace, scran’s findMarkers, and Seurat’s FindAllMarkers, respectively, based on spatial clusters computed via BayesSpace and StLearn.

Figure 4.

Expression plot, for four SVGs detected with DESpace individual cluster test, on the real LIBD dataset (sample 151673). SVGs were identifying by selecting high and low expression in white matter (genes MOBP and ENC1, respectively), and high and low abundance in layer 3 (genes HOPX and HS3ST4, respectively). Lines highlight the cluster being tested.

Figure 5.

TPR versus FDR for SVG detections in the multiple sample simulation. Rows and columns refer to the anchor data used in the simulation, and to the SV profiles, respectively. Note that TPRs are lower than in the individual simulation (Fig. 3), because we have simulated slightly weaker spatial patterns here (see Supplementary Details).

Figure 6.

Jaccard index on the LIBD and melanoma datasets. For each method, the index is measured on the sets of top SVGs reported across replicates, and represents how coherent results are between samples: higher values indicate greater coherency.