Supplementary Figure 8: Comparison of SpatialDE to alternative measures of expression heterogeneity in the breast cancer tissue.

(A) Comparison of adjusted negative log P-values from the SpatialDE test (y-axis) versus commonly used statistics for expression heterogeneity (x-axis) - Upper left: Mean, Upper right: Variance, Lower left: CV2 (squared coefficient of variation), Lower right: Dropout rate (fraction of cells/samples a gene is not detected in). Random selection of significant SV genes highlighted in red for context. No dependence between SpatialDE significance levels and expression level (mean) or variance was observed. Statistics calculated for 12,856 genes using 250 “spots”. (B) Comparison of significance of SV for genes identified by SpatialDE versus commonly used strategies for defining highly variable genes based on regression models between summary statistics: Relation with CV2 (Upper) or Variance (Middle), or with dropout fraction (Bottom). The rightmost column of plots show residuals compared with the SpatialDE significance; polynomial regression for CV2 and Variance, logistic regression for dropout rate. Significant SV genes as identified by SpatialDE (FDR<0.05, Q-value adjusted) are shown in grey. Other, non-significant genes are shown in solid black. The SV genes significance are orthogonal to HVG measures, indicating that spatial variation is different from general variability. Statistics calculated for 12,856 genes using 250 “spots”.