Fig. 3

Validation of EPISCORE-derived DNAm reference matrix for lung. a Heatmaps representing the scRNA-Seq reference matrix for lung (top) and the corresponding inferred DNAm reference matrix (bottom), defined over the 4 main cell types found in lung tissue: Epithelial (Epi), endothelial (Endo), fibroblasts (Fib), and immune cells (IC) and over the common set of marker genes. Note that in the case of the DNAm reference matrix, each column represents the average DNAm within 200 bp of the transcription start site of the gene. The colorbar at the bottom displays the weight of the corresponding gene with weights close to 1 indicating genes for which differential DNAm is highly associated with differential expression across the 4 cell types. Examples of genes with weights > 0.8 and which are known markers for their respective cell types are shown. b Validation of the inferred DNAm reference matrix in a using 100 in silico generated mixtures of DNAm 450k profiles representing purified epithelial, endothelial, fibroblast, and immune cells derived from independent studies. The first four panels display scatterplots of the estimated cell type fraction (x-axis) vs. the true fraction (y-axis) for each main cell type, with the Pearson correlation coefficient (PCC), as shown. The last panel displays density plots of the PCC for each cell type (as estimated over 100 in silico mixtures), as obtained from 100 different runs