Fig. 7: Subtyping anomalous tissue domains (ATDs) across multiple human breast cancer 10x Visium datasets with both shared and dataset-specific domain subtypes.

a The ATDs include the CIS, IC, and adipose tissue (AT) subtypes. As shown in the ground truth panel of spatial maps, the CIS (in orange) domain is shared between the two experimental datasets (10x-hBC-H1 and 10x-hBC-G2), while the IC (in green) and AT (in red) domains are unique to 10x-hBC-G2 and 10x-hBC-H1, respectively. Each of the tested methods (STANDS and six benchmark methods) sequentially identifies, aligns, and subtypes anomalous spots, with subtypes indicated in orange (CIS), green (IC) and red (AT) in methods’ respective panels of spatial maps. b Sankey diagrams detail the correspondence between the ground truth and the anomaly subtyping results from Fig. 7a. c The left scatterplot illustrates the Macro-F1 (y-axis) vs NMI (x-axis) scores of the methods in identifying and subtyping anomalous spots, respectively. The right scatterplot illustrates methods’ multi-SGD_degree (x-axis) vs multi-SGD_cc (y-axis) scores. Lower scores in both multi-SGD metrics indicate not only more accurate labeling but also that the subtyping results are more consistent with the ground truth in their spatial arrangement (see “Evaluation metrics” section in Methods). d The heatmaps of multi-SGD_degree (top row) and multi-SGD_cc (bottom row) scores of the methods. In each heatmap, the color intensity of cell \({c}_{i,j}\) inversely correlates with the SGD score (either degree or cc) calculated with the presumption that the method-generated subtype \(j\) corresponds to the true subtype \(i\). Cells on the leading diagonal indicate the optimal correspondence between the method’s subtyping results and the true subdomain labels (see “Evaluation metrics” section in Methods). Therefore, a more pronounced color pattern along a heatmap’s leading diagonal signifies more accurate subtyping. Source data are provided in this paper.