Fig. 4: Deconvolution of immune cell types on Illumina EPIC array.
From: Benchmarking of methods for DNA methylome deconvolution
a Matrix of marker CpGs (n = 600) used for building immune cell methylation reference of EPIC data. Samples for six cell types were included: neutrophil (n = 6), natural killer cell (n = 6), B-cell (n = 6), CD4+ T-cell (n = 7), CD8+ T-cell (n = 6) and monocyte (n = 6). b Deconvolution accuracy represented as boxplots showing accuracy scores for the different deconvolution methods, normalization methods, and cell types on 200 in silico mixtures. Black diamond shapes represent median values, and colors represent cell types. The boxplots present median values and quartiles, whiskers the minimum and maximum values, and dots the individual data points. P-values were determined using two-tailed FDR-adjusted Dunn’s tests. *P < 0.05, **P < 0.01, ***P < 0.001. c Performance of deconvolution on 200 in silico mixtures for all algorithm-normalization combinations, represented as circles. Spearman’s R2 is represented by color, root mean squared error (RMSE) is represented by size. Rows show deconvolution algorithms, columns show normalization methods. d Boxplots showing RMSE values of in silico (n = 200) and in vitro (n = 12) datasets for CD8+ T-cells, natural killer cells, and all cell types on CpGs selected for 450 K and EPIC array data. The boxplots present median values and quartiles, whiskers the minimum and maximum values, and dots the individual data points. e Scatter plots of 200 in silico mixtures showing true proportions (x-axis) and predicted proportions (y-axis) in percentages for the best-performing (left-upper) and worst-performing (right-lower) algorithm-normalization combinations on in silico EPIC data. R2 and p-values were calculated using Spearman’s rank correlation test. f Scatter plots of in vitro data (n = 12) showing true proportions (x-axis) and predicted proportions (y-axis) in percentages for the best-performing (left-upper) and worst-performing (right-lower) algorithm-normalization combinations on in silico EPIC data. R2 and p-values were calculated using Spearman’s rank correlation test. ‘X’ symbol represents missing values. Source data are provided as a Source Data file. Exact p-values are added in the Source Data file.
