Fig. 4

Layers of spatial gene expression variability. a Schematic representation of the subsequent steps needed to dissect the different layers of spatial gene expression variability. The original cell locations, a spatial grid, or a spatial network is required to identify individual genes with spatial coherent expression patterns. Those spatial genes can then be used as input to compute continuous spatial co-expression patterns or to find discrete spatial domains with HMRF. b–d Spatial gene expression analysis of the seqFISH+ somatosensory cortex dataset (distance unit = 1 pixel, 1 pixel ≈ 103 nm). b Examples of identified spatial genes within the somatosensory multi-layered cortex. The outer layers are on the left, while more inner layers are on the right. c Overlap between the top 1000 spatial genes identified from the 5 methods implemented in Giotto. d Visualization of spatial domains identified by the HMRF model. The layered anatomical structure (L1–6) of the somatosensory cortex is indicated on top. e, f Spatial gene expression analysis of the Visium kidney dataset (distance unit = 1 pixel, 1 pixel ≈ 1.46 μm). e Heatmap showing the spatial gene co-expression results. Identified spatial co-expression modules are indicated with different colors on top. f Metagene visualizations for all the identified spatial gene co-expression modules from e, g Selected gene visualizations for each identified spatial metagene in e and f