Fig. 4: SpaGFT reveals the follicle heterogeneity on Tonsil CODEX data.

a A 49-plex CODEX data was generated from human tonsil tissue at a 0.37 μm/pixel resolution. Six FOVs were selected based on their varying tissue microenvironment and cellular organization. b Cell phenotype maps for each of the six FOVs, depicting the cellular composition and organization. c The results showed the characterization FTUs based on the gradient pixel-level images for A6. The heatmap depicts the SSIM score, where a higher score corresponds to a lighter color and greater structural similarity. d A heatmap showcasing the protein expression of each FTUs represented by the six SVP clusters, which were identified as FTUs resembling secondary follicles. The values in the heatmap are scaled by z-scores of protein expression. e–h Overlays of CODEX images for SVPs for FOVs 1, 3, 4, and 6, respectively. i. Spatial maps depicting the patterns of secondary follicle FTUs from six FOVs. Dash rectangles indicate the identified follicle regions. Note that panels d to h are ordered by FOV 1, 2, 3, 4, 5, and 6. j Cell phenotype maps of the FTUs identified in (i). k Barplots depicting the cell components of the identified FTU in (i). The cell type colors were depicted in (b). l The graph network depicting the spatial proximity of the top 5 abundant cell types in the FTU identified in i, as calculated by \(\frac{1}{1+d}\), where d represents the average distance between any two cell types. m Dumbbell plots indicated significant cell–cell interaction among B cells and others. If the observed distance is significantly smaller than the expected distance, the two cell types tend to be contacted and interact. Line length represents relative distances, subtracting the expected distance from the observed distance. An empirical permutation test was used to calculate the p-value, and the point size was scaled using an adjusted p-value.