Supplementary Figure 7: Comparison to differential expression analysis using unsupervised clustering of spots.

(A) Principal component analysis using genome-wide expression profiles of individual “spots” from the spatial transcriptomics breast cancer data, color coded by cluster membership for K=4 clusters (using Bayesian Gaussian Mixture Modelling). (B) Bayesian Gaussian Mixture Model cluster probabilities, discretizing the 250 spatial breast cancer “spots” into four clusters (analysis ignores spatial structure). (C) Visualization of cluster membership in the original tissue context. (D) Scatter plot of negative log P-values from an ANOVA test between clusters (x-axis) versus negative log P-values of spatial variation from SpatialDE (y-axis). 83 genes were identified as significantly variable by both approaches (FDR<0.05, Q-value adjusted); 32 genes are significant only in the SpatialDE test, among them immune genes. (E) Histogram of the fitted length scales for SV genes detected by both approaches (blue) and SV genes exclusively detected by SpatialDE (orange). Genes that were only detected by SpatialDE were associated with smaller length scales, indicating localized expression patterns.