Fig. 4: Results on the human breast cancer dataset.

a H& E image of the human breast cancer dataset, manual annotation, and clustering results of HAST and baseline methods on this slice measured by ARI and FMI. Colors are independently assigned for each method. b Differential gene expression (DGE) analysis of Cluster 4 vs. other clusters (left) and Cluster 16 vs. other clusters (right). Each point represents a gene, with the vertical axis indicating -log10 of the p-value and the horizontal axis representing the log2FoldChange (log2FC). P-values were from the two-sided Wilcoxon rank-sum test. Significance thresholds were set at ∣log2FC∣ > 0.5 and p-value  < 0.05. c Violin plots of CAF marker genes (TIMP1, COL1A2, DCN) in Cluster 4 vs. other clusters. The vertical axis represents gene expression levels. Each violin indicates the expression distribution of a specific gene, and the width indicates the frequency. The central white line represents the median, the thick black bar within each violin is the interquartile spacing, and the whisker line extends from the 25th and 75th percentiles to 1.5 × interquartile range. d GO analysis of Cluster 4 vs. other clusters. The vertical axis represents GO terms, while the horizontal axis represents -log10 of the p-value. GO enrichment analysis was conducted using a one-sided hypergeometric test, with p-values adjusted for multiple comparisons using the Benjamini-Hochberg method.