Fig. 4

Reproducibility of SVG detection tools with down-sampling of the data. a Schematic of the experimental approach to determine the interdependency of genes in the calculation of SVG statistics. For the down-sampling, 50% of the genes were randomly down-sampled while keeping the number of spots equal. b Boxplots of the Spearman correlation results performed on all datasets colored by SVG method and ordered by increasing mean correlation coefficient. c, d As in a, b but for the investigation of the reproducibility of SVG methods with increased sparsity in spatial spots. The datasets were randomly down-sampled to 80% of the total number of spots. e Venn diagram illustrates the proportions of uniquely identified SVGs as described below in f. Boxplots of the proportion of false positive SVGs calculated as the proportion of SVGs uniquely identified in the down-sampled data divided by all the significant SVGs identified in the down-sampled data