Fig. 2: Compositional and transcriptional remodeling of the immune microenvironment in Hodgkin lymphoma.

a UMAP embedding of immune cells colored by cell type annotation (n = 277,911 nuclei from 15 subjects). b compositional analysis of immune cells in Hodgkin lymphoma (HL) versus reactive lymphoid tissue (RLN) samples. c unsupervised consensus tensor factorization reveals a shared HL-associated myeloid cell program. d compositional analysis of myeloid subsets across HL and RLN samples. e UMAP embedding of stromal cells colored by cell type annotation (n = 23,772 nuclei from 15 subjects). f compositional analysis of stromal cells in HL versus RLN samples. g, consensus tensor factorization reveals a distinct activated pro-fibrotic, myofibroblast-like fibroblastic reticular cell program. For all these plots (including b, d, and f), n = 15 (10 HL and 5 RLN) samples from independent subjects (biological replicates). For the boxplots (b, d, f), boxes represent the interquartile range (25th − 75th percentiles) with the median shown as a central line; whiskers extend to the most extreme data points within 1.5 × IQR of the quartiles. All data points are shown in the boxplots. Statistical significance of the compositional analysis (b, d, f) was assessed using the scCODA framework, which estimates posterior inclusion probabilities for each covariate-cell type effect and applies a Bayesian FDR threshold to determine significance. Effects with posterior inclusion probabilities exceeding the cutoff corresponding to FDR < 0.20 are marked with *. Tests are two-sided, as the model estimates both enrichment and depletion.