Fig. 3

CoSTA Analysis of Slide-seq data. A Shuffling test to disrupt spatial patterns. Left panel: The first row shows the three original spatial expression patterns of three example genes. Images in the second row are spatial patterns after shuffling (all images shuffled in the same way so that pixel-level overlap is preserved while spatial neighbor relationships are broken- see Methods and Additional file 12: Fig. S12 for shuffling approach details). Right panel: CoSTA-derived distances between 9 randomly selected genes and Prdx5. Genes are ordered based on how close they are to Prdx5 using spatial features extracted by CoSTA from true gene matrices (left to right: closest to farthest). Shuffled gene matrices are forwarded through CoSTA, and distances between gene pairs are subtracted from the unshuffled distances. Each point represents distance change for one shuffling (100 shufflings total). Red line at 0 indicates no change in distance would be observed using overlap calculations. B The number of overlapped gene neighbors of Vim, Ctsd, and Gfap before and after each weight updating across all training epochs (30 nearest neighbors considered, see Additional file 5: Fig. S5 for different size neighbor sets). Results shown for two experiments: 3 days (blue) or 2 weeks (red) after brain injury. C Overlap of CoSTA, Spatial DE and SPARK genes called SE and correlated with Vim, Ctsd, and Gfap in the 2 weeks after injury dataset. Yellow = SPARK correlated genes also SE by CoSTA. Blue = SPARK correlated genes not SE by CoSTA. Red cross-hatching = Proportion of each category also identified by SpatialDE. Below, Gene Ontology enrichment (Panther) of genes that overlap between SPARK, SpatialDE AND CoSTA (left) and those that overlap between SPARK and Spatial DE but NOT CoSTA (Right). D GO term enrichment in the 2 weeks after injury Vim, Ctsd, and Gfap correlated gene sets from different approaches for biologically relevant functions identified by the original Slide-seq analysis. Quantified along the axis is the fraction of genes in each method’s correlated gene list that are annotated with the given GO term