Fig. 6: Mapping and clustering of IMC samples.

a Scores of the best-matching patient pairs across 100 sampled breast cancer datasets. We obtained higher mapping scores in metrics linked to the cell state continuum (cell, niche, territory). For each best-mapped patient pair, we also calculated the correspondence between their respective clinical metrics. 80% of samples share at least 3 of the 5 clinical metrics observed (PAM−50, ERRBB_2 positive, Grade, ER Status, and deathBreast). b Across the 5 clinical metrics, we observed that ERBB2_pos is the metric that is the most accurately recovered (89%) while PAM50 (cancer subtype) is the most poorly predicted (49%). c Clustering using ER-negative breast cancer samples. The distinction of these samples is mainly driven by cell label and niche composition, which contrasts with the continuous cell state metrics. Clusters 9 and 10 – for instance – show little to no difference in cell, niche, and territory similarity in contrast to the stark difference in cell label and niche composition scores. d We highlight the clinical metrics associated with each cluster. We use the term Boolean to define True (1) or False (0). Cluster 9 is predominantly characterized by the HER2 cancer subtype, and cluster 10 is characterized by the Basal cancer subtype. Source data are provided as a Source Data file.