Extended Data Fig. 5: Metaclustering and cluster matching across cohorts.

a, Heat map showing z-scored mean marker expression of single-cell phenotypic clusters identified by PhenoGraph (Fig. 1) with colours on the colour bar and hierarchical clustering indicating the corresponding metacluster. Red stars on the hierarchical clustering tree indicate subgroups that robustly reappear as separate groups using multiscale bootstrap resampling (P < 0.05 (R function pvclust)). b, Examples of untransformed distributions of cluster marker expressions that differ between metaclusters. c, Heat maps showing the z-scored mean marker expression or distance to the tumour–stroma interface for each metacluster defined in the cohort of 281 patients from University Hospital Basel and each matched PhenoGraph cluster from the multicore cohort of 71 patients from University Hospital Zurich. PhenoGraph clusters of the Zurich cohort were matched to the metaclusters of the Basel cohort on the basis of the Pearson correlation of the mean marker expression (Methods).