Optimized Cellular Deconvolution of Skin Tape RNA-seq Reveals Cell Remodeling in Infant Atopic Dermatitis
Recommended Citation
Fluke K, Jackson N, Everman J, Diener K, Pruesse E, Keet C, Kim E, Shreffler W, Hui-Beckman J, Doan D, Sicherer S, Bunyavanich S, Hershey GK, Wood R, Singh A, Sindher S, Bacharier L, Johnson C, Anvari S, Togias A, Fulkerson P, Yeh E, Goleva E, Leung D, Seibold M, Moore C. Optimized Cellular Deconvolution of Skin Tape RNA-seq Reveals Cell Remodeling in Infant Atopic Dermatitis. J Allergy Clin Immunol 2026; 157(2):AB424.
Document Type
Conference Proceeding
Publication Date
2-10-2026
Publication Title
J Allergy Clin Immunol
Keywords
atopic dermatitis, benchmarking, cell differentiation, conference abstract, controlled study, correlation coefficient, deconvolution, epidermis, female, fibroblast, food allergy, human, human tissue, infant, keratinocyte, Langerhans cell, male, RNA sequencing, root mean squared error, scar formation, scar tissue, single cell RNA seq, skin, skin biopsy, T lymphocyte
Abstract
Rationale: Studies investigating molecular changes in atopic dermatitis (AD) skin have been limited by scarring biopsies, while minimally invasive skin tape strippings (STS) have enabled large-scale investigations into AD pathobiology. Moreover, computational deconvolution of bulk RNA-seq from STS samples is a promising strategy to investigate the cellular changes that accompany AD, but this methodology has yet to be optimized or applied to STS RNA-seq data. Methods: We generated scRNA-seq data on 18 skin biopsies (6 lesional, 5 non-lesional, 6 healthy) to build an STS-appropriate reference. We benchmarked deconvolution methods on simulated bulk RNA-seq cell mixtures using Pearson’s correlation coefficients, Lin’s concordance correlation coefficients (CCC), and root mean squared error. We then deconvolved non-lesional STS RNA-seq from SUNBEAM cohort infants (n=87) with and without AD and food allergy (FA), comparing inferred cell type proportions between disease groups. Results: InstaPrism yielded the highest accuracy (median CCC=0.85), followed by DISSET, CibesortX, Scaden, and Bisque. Estimates were most reliable for mature/terminally differentiated keratinocytes and fibroblasts, and least accurate for T cells. Separating suprabasal, early transitioning, and intermediate/spinous keratinocyte states reduced accuracy, whereas combining these into a single, mid-keratinocyte compartment markedly improved concordance. In deconvolved infant STS data, AD and AD+FA samples showed increased Langerhans cells and reduced mature keratinocytes compared to healthy controls. Conclusions: Accurate deconvolution of STS requires method benchmarking and a reference tailored to superficial epidermis. With an optimized pipeline, deconvolution of STS RNA-seq data can be used to resolve disease-associated shifts in skin composition, enabling scalable profiling of AD cellular pathobiology.
Volume
157
Issue
2
First Page
AB424
