Integrative Single-Cell Transcriptomic Analysis Identifies Candidate Endothelial Activation Signatures in Atopic...

Background: Endothelial activation is a recognized histological feature of atopic dermatitis (AD), but its transcriptional programs remain incompletely characterized. Methods: We reanalyzed public single-cell RNA-seq data from the skin of five patients with AD and four healthy donors using AD-associated gene set scoring, CellChat, endothelial hdWGCNA and pseudotime ordering, and examined candidate genes with eight machine learning classifiers and SHAP in public bulk transcriptomes of lesional and non-lesional AD skin. Results: Inferred intercellular interactions were more numerous in AD (1348 vs. 1111), with COLLAGEN, CD99 and MIF showing the largest gains in communication strength. Differences in AD-associated gene set scores between AD and healthy endothelial cells were small and inconsistent between scoring approaches. Among seven endothelial modules, Endo7, an NF-κB-responsive immediate-early module containing SELE, ICAM1, IRF1 and NFKBIA, overlapped most strongly with the AD gene set (10 genes; approximately 30-fold enrichment). AD endothelial cells were concentrated at intermediate pseudotime positions (Cohen's d = 0.334). In bulk skin, classifiers based on 14 LASSO-selected genes separated lesional from non-lesional AD skin in a small test set (AUC 0.70-0.98; likely optimistic estimates, as feature selection preceded the training-test split and paired samples from the same patients may span both partitions), with SELE, TOP2A and IRF1 ranked highest by SHAP. Conclusions: This hypothesis-generating analysis identifies AD-associated endothelial activation signatures and candidate genes that require confirmation in independent, clinically characterized cohorts and experimental studies.