(A) Cell type level label transfer results, where RA synovium scRNA-Seq was used as the reference and JIA synovium Xenium data as the query. RA-derived annotations were transferred to individual JIA cells using a k-nearest neighbors (kNN) approach in the integrated latent space. (B) Match rate between transferred RA-synovium cell types and our original JIA-synovium-based annotations. Matrix shows percentage overlap between each transferred and original broad cell type. (C) Cell clustering in the integrated latent space. UMAP visualizations of both JIA (Xenium) and RA (scRNA-Seq) cells are colored by newly defined shared clusters. (D) CNA identifies integrated-space clusters that are expanded or depleted in JIA synovium relative to RA synovium. Sex and disease duration were adjusted for, while age was excluded from covariates due to its collinearity with disease status, reflecting the distinct age of onset between RA and JIA. Left: Spatial visualization of cells colored by neighborhood correlation (red, expanded in JIA; blue, depleted). Middle: Violin plots showing distribution of neighborhood correlations for each integrated cluster. Only clusters with cells either of positive or negative correlation > 80% are shown. Right: Bar plots showing cell-type composition of each cluster in RA (left) and JIA (right) synovium. For each cluster, the most abundant original cell type (from either RA or JIA) is annotated with text in each bar. Center bar plot showing relative proportion of cells derived from JIA versus RA within each cluster (orange, JIA; blue, RA). (E) Box plots showing the lineage-normalized proportion of cell clusters across individual samples, stratified by condition (JIA versus RA synovium). P values were calculated using Wilcoxon rank-sum tests. Box plots showing the median, interquartile range, and 1.5× interquartile range (IQR) whiskers.