Fig. 1

Performance assessment of signature matrix DerM22 for deconvolution of gene expression data from skin biopsies. a The accuracy of DerM22 is analyzed by comparing the mean deconvolution results (blue) to mean flow cytometry data (red) for samples of healthy subjects (triangle), and the lesional skin of psoriasis patients (square). For the flow cytometry data and CIBERSORT results, the standard error is shown for all cell subsets except for the flow cytometry-derived psoriasis pDCs values given that it was not reported in the original study. Also, the cross-platform bias is explored by the comparison of the deconvolution results (blue) from microarray data acquired in the same platform as DerM22, i.e. Affymetrix HG-U133 Plus 2.0, to data obtained using a different platform (green), i.e. Affymetrix HG-U133A. b The performance of DerM22 is also compared to immunohistochemistry results from three different studies, i.e. GSE36842 (yellow), GSE27887 (black), GSE32924 (white), for which expression data is also available. This is done by deriving the ratio of average myeloid dendritic cells (CD11c), macrophages (CD206), and CD8⁺ T cells in lesional to non-lesional skin, or in treated to untreated lesional skin of atopic dermatitis (AD). The plotted ratios correspond to the ratios of the average for each subgroup. c The cross-platform bias is further analyzed by deconvolving expression data from 25 different isolated cell types acquired with Affymetrix HG-U133A. The estimated fractions of each cell type are presented in a heatmap, ranging from 0% (black) to 100% (bright blue). d-f The consistency of the mean results (+/− standard deviation) across independent datasets of skin in health and disease is examined for CD4⁺ T cells, plasmacytoid dendritic cells (pDCs), and keratinocytes. For results on other cell types see Additional file 5: Fig. S2